Changelog

• Fixed activation errors across vaults on the same desktop device and restored existing device authorization after sign-in. Improved Home loading and workspace restoration.
• Clarified the scope of the MIT-licensed base and commercially licensed PRO components, preserving existing grants and third-party rights. See LICENSE.

After upgrading, open the vault where you previously activated this device once to migrate its existing credentials. For manual upgrades, replace only the three plugin files below and keep your settings and data.

View on GitHub ↗

• Added official Obsidian CLI support for opening notes and tabs, working with Bases, appending daily notes, reading or restoring local history, and finding and managing plugins through chat.
• Fixed default daily-note paths and Skills disappearing from the / menu after typing or reopening it. Chinese and English names remain searchable.

CLI features require the official Obsidian CLI to be installed and enabled on desktop. Writes and plugin management follow the confirmation flow. Plugin search downloads the official public directory from GitHub and filters it locally; notes and search terms are not uploaded to the directory service. To upgrade, replace the three plugin files below and keep your settings and data.

View on GitHub ↗

Fixed Codex authorization and chat compatibility. Fixed compatibility issues with certain Codex request parameters, authorization tokens, and streamed responses. Connection tests now use lower reasoning effort and wait up to 60 seconds. Request rejection, unavailable models, usage limits, network failures, and expired authorization now produce distinct messages, reducing misleading response-format errors and requests to sign in again.

Fixed Codex chat connectivity. Fixed an issue where desktop OpenAI Codex Beta could report a network interruption when sending messages after authorization and model setup. All other features remain as in 2.5.0.

Search outside your vault when you need sources, and archive conversations when you are done with them.

2.5.0 brings Tavily web search and automatic conversation archiving to desktop. Both are off by default, so you choose when to enable them.

• Search the web and follow sources. Once Tavily is configured, your current model can search when a question calls for it, read snippets, and cite the returned links. You can keep your existing chat model.
• Configure models and tools separately. The API Provider page now has model API and tool API tabs. Tavily includes key visibility controls, a connection test, and its own switch. A successful test does not enable search automatically.
• Organize older conversations by age. Choose 7, 14, 30, or 90 days in Review settings. Conversations with no updates beyond that period are archived automatically. Conversations being viewed, running, queued, or holding an unsent draft are skipped. Archived conversations can be restored; nothing is deleted, and no model service is required.
• Keep settings usable in narrow windows. Tavily descriptions and the search switch now stay aligned in smaller settings windows.

Get started

Web search: open Settings → EchoInk Agent → API Provider → tool 工具 (Tools), enter your own Tavily API key, test the connection, and enable web search. Each connection test makes one basic search and consumes 1 credit. Normal searches also use basic and return up to 5 results. A separate Tavily account and key are required; quotas and pricing follow its current policy. Switch changes apply to the next turn of an existing conversation.

Automatic archiving: choose a period under Review → Management → Automatic archiving. Checks run only while the plugin is active. Restore conversations from Archived conversations. Conversation pinning is not included. Some search activity summaries may still use a generic tool title; web results and source links remain available.

The new search and archiving features are desktop-only. Mobile retains its existing Beta features, and real iOS and Android device validation remains incomplete. Search queries are sent to Tavily; returned snippets and links are passed to your current model for its answer. Use search according to what you are comfortable sharing.

Download and upgrade

Requires Obsidian 1.11.4 or later. Download main.js, manifest.json, and styles.css from this release and place them in the vault's .obsidian/plugins/codex-echoink/ folder. Keep existing data.json and other data files, then re-enable the plugin. You can also check for updates in Obsidian Community plugins once its directory picks up this version.

View on GitHub ↗

Fixed Codex chat connectivity. Fixed an issue where desktop OpenAI Codex Beta could report a network interruption when sending messages after authorization and model setup. All other features remain as in 2.5.0.

Search outside your vault when you need sources, and archive conversations when you are done with them.

2.5.0 brings Tavily web search and automatic conversation archiving to desktop. Both are off by default, so you choose when to enable them.

• Search the web and follow sources. Once Tavily is configured, your current model can search when a question calls for it, read snippets, and cite the returned links. You can keep your existing chat model.
• Configure models and tools separately. The API Provider page now has model API and tool API tabs. Tavily includes key visibility controls, a connection test, and its own switch. A successful test does not enable search automatically.
• Organize older conversations by age. Choose 7, 14, 30, or 90 days in Review settings. Conversations with no updates beyond that period are archived automatically. Conversations being viewed, running, queued, or holding an unsent draft are skipped. Archived conversations can be restored; nothing is deleted, and no model service is required.
• Keep settings usable in narrow windows. Tavily descriptions and the search switch now stay aligned in smaller settings windows.

Get started

Web search: open Settings → EchoInk Agent → API Provider → tool 工具 (Tools), enter your own Tavily API key, test the connection, and enable web search. Each connection test makes one basic search and consumes 1 credit. Normal searches also use basic and return up to 5 results. A separate Tavily account and key are required; quotas and pricing follow its current policy. Switch changes apply to the next turn of an existing conversation.

Automatic archiving: choose a period under Review → Management → Automatic archiving. Checks run only while the plugin is active. Restore conversations from Archived conversations. Conversation pinning is not included. Some search activity summaries may still use a generic tool title; web results and source links remain available.

The new search and archiving features are desktop-only. Mobile retains its existing Beta features, and real iOS and Android device validation remains incomplete. Search queries are sent to Tavily; returned snippets and links are passed to your current model for its answer. Use search according to what you are comfortable sharing.

Download and upgrade

Requires Obsidian 1.11.4 or later. Download main.js, manifest.json, and styles.css from this release and place them in the vault's .obsidian/plugins/codex-echoink/ folder. Keep existing data.json and other data files, then re-enable the plugin. You can also check for updates in Obsidian Community plugins once its directory picks up this version.

View on GitHub ↗

Search outside your vault when you need sources, and archive conversations when you are done with them.

2.5.0 brings Tavily web search and automatic conversation archiving to desktop. Both are off by default, so you choose when to enable them.

• Search the web and follow sources. Once Tavily is configured, your current model can search when a question calls for it, read snippets, and cite the returned links. You can keep your existing chat model.
• Configure models and tools separately. The API Provider page now has model API and tool API tabs. Tavily includes key visibility controls, a connection test, and its own switch. A successful test does not enable search automatically.
• Organize older conversations by age. Choose 7, 14, 30, or 90 days in Review settings. Conversations with no updates beyond that period are archived automatically. Conversations being viewed, running, queued, or holding an unsent draft are skipped. Archived conversations can be restored; nothing is deleted, and no model service is required.
• Keep settings usable in narrow windows. Tavily descriptions and the search switch now stay aligned in smaller settings windows.

Get started

Web search: open Settings → EchoInk Agent → API Provider → tool 工具 (Tools), enter your own Tavily API key, test the connection, and enable web search. Each connection test makes one basic search and consumes 1 credit. Normal searches also use basic and return up to 5 results. A separate Tavily account and key are required; quotas and pricing follow its current policy. Switch changes apply to the next turn of an existing conversation.

Automatic archiving: choose a period under Review → Management → Automatic archiving. Checks run only while the plugin is active. Restore conversations from Archived conversations. Conversation pinning is not included. Some search activity summaries may still use a generic tool title; web results and source links remain available.

The new search and archiving features are desktop-only. Mobile retains its existing Beta features, and real iOS and Android device validation remains incomplete. Search queries are sent to Tavily; returned snippets and links are passed to your current model for its answer. Use search according to what you are comfortable sharing.

Download and upgrade

Requires Obsidian 1.11.4 or later. Download main.js, manifest.json, and styles.css from this release and place them in the vault's .obsidian/plugins/codex-echoink/ folder. Keep existing data.json and other data files, then re-enable the plugin. You can also check for updates in Obsidian Community plugins once its directory picks up this version.

View on GitHub ↗

Take your thoughts with you.

2.4.1 introduces Mobile Beta. The same plugin package opens the mobile interface on a phone and keeps the existing desktop interface and features on a computer.

• Chat on your phone. Configure an API key, endpoint, and model to start asking questions. OpenAI Chat Completions and Responses are supported. Replies currently appear after the complete response arrives.
• Read notes and create Markdown. Find and read notes in the current vault, then ask EchoInk to turn them into a new Markdown file. Existing files with the same name are not overwritten, and read-only access prevents note creation.
• Keep useful context. Save and recall local memory in mobile conversations, or view, edit, and forget individual entries. Turning long-term memory off stops recall and writes. Existing local history and memory can still be managed without a configured model.
• Customize the welcome. Change the empty conversation's title and greeting in General settings; edits save automatically. Disabling customization or leaving a field blank restores the default wording. Turning it off keeps your saved text.

Use and upgrade

Download main.js, manifest.json, and styles.css from this release and place them in the vault's .obsidian/plugins/codex-echoink/ folder. Keep existing data.json and mobile/, then re-enable the plugin. You can also check for updates in Obsidian Community plugins once its directory has picked up this version.

On mobile, run 打开移动对话 (Open mobile chat), then configure a model under Settings → API Provider → 模型与提供商 (Models and providers). Manage local memory under 复盘 → 记忆 (Review → Memory) and customize the welcome under 基础设置 (General). The mobile interface is currently in Chinese. Desktop users can continue with their existing workflow.

Requires Obsidian 1.11.4 or later. Mobile Beta prioritizes iOS. Android uses the same mobile entry and is available for experimental testing; real iOS and Android device validation is not complete. Desktop retains its existing capabilities.

Mobile does not yet include cross-device memory sync, background tasks, CLI, attachments, or the complete desktop Skills/MCP features. Mobile and desktop conversations and memory are stored separately. The phone screen in the cover comes from a Demo and is illustrative; not every module shown is connected.

View on GitHub ↗

Follow the connections between your notes.

2.4.0 lets Ask continue reading through related notes and gives maintained knowledge more context about when it applies.

• Read related material as needed. After finding a note, the model can inspect its outgoing links and backlinks, then choose what to read next. It decides whether to find more material or answer based on the question; it does not need to open every link.
• Keep time periods and conditions. /maintain can add subjects, time periods, source types, applicability conditions, and relationship descriptions to help distinguish projects, versions, and budgeted versus actual figures. Existing notes remain usable without a fixed set of required fields.
• Distinguish what has been read. Delivered passages are tracked separately from leads still to explore. Partially read notes, failed reads, and incomplete results are no longer treated as fully read. A missing source does not prevent the model from continuing with other available material.
• Clearer failures and consistent presentation. Missing files or changed paths now show the specific cause when known, and failed operations are no longer labeled as complete. Older records without a detailed cause say so. Failure cards use font sizes, line heights, and spacing consistent with surrounding process entries.

Use and upgrade

Enter /ask in a conversation and ask, for example, “How much did this project actually cost, and how did that compare with the budget?” The model can consult related notes as needed. When organizing new material with /maintain, you can ask it to preserve time periods and applicability conditions. Upgrading does not require vault reinitialization or model reconfiguration.

Related reading provides more evidence for a judgment. Check source passages and calculations for conclusions involving amounts or ratios; the model can still misunderstand conditions or make inaccurate inferences.

Check for updates in Obsidian's Community plugins page, or download main.js, manifest.json, and styles.css from this release, place them in the vault's .obsidian/plugins/codex-echoink/ folder, and re-enable the plugin. Keep your own data.json.

Requires Obsidian Desktop 1.11.4 or later. Mobile is not supported. Related reading and failure feedback have been validated in the macOS app; corresponding Windows/Linux interactions have not received real-app validation. The cover is illustrative.

View on GitHub ↗

Lighter to download. Smoother to use.

2.3.2 focuses on a smaller plugin bundle and better dropdown interactions in settings and chat.

• Main file reduced by about 46.4%. main.js shrinks from about 9.27 MB in 2.3.1 to about 4.97 MB (4.74 MiB), below the 5 MiB per-file limit of Obsidian Sync Standard. PDF reading, image processing, full Chinese pinyin search, and existing model capabilities are preserved.
• Click outside to cancel. Settings dropdowns and chat model/parameter menus close when you click a blank area or elsewhere, keeping the previous selection. Hover highlights an option; clicking confirms it. This also fixes menu opening and dismissal in separate windows.
• Brand icons for the current model. The selected value and options under EchoInk current model now use the same icons as the Provider page, making services such as DeepSeek easier to recognize.

Use and upgrade

Keep using your existing settings and models after upgrading; vault reinitialization is not required. Open Settings → EchoInk Agent → Knowledge to see the current model's brand icon. Click outside settings dropdowns and chat parameter menus to cancel.

Check for updates in Obsidian's Community plugins page, or download main.js, manifest.json, and styles.css from this release, place them in the vault's .obsidian/plugins/codex-echoink/ folder, and re-enable the plugin. Keep your own data.json. Whether plugins are synced still depends on your sync settings.

Requires Obsidian Desktop 1.11.4 or later. Mobile is not supported. Window interactions have not received real-app validation on Windows/Linux for this release. The cover is illustrative.

View on GitHub ↗

Readable folders. Clearer organization progress.

2.3.1 focuses on everyday improvements to knowledge organization, to-do statistics, and action feedback.

• Bilingual folder names: Initialization and maintenance use Chinese and English category and folder names. In existing vaults, select Optimize names to rename English root folders and Wiki category folders on disk—for example, raw becomes 原始资料（raw）. Existing bilingual folders are reused.
• Restore vault: Initialization saves a record of the original folder layout. Restore moves recorded files back to their original locations while keeping their latest content and new files. Conflicts are preserved and explained. This restores locations, not earlier versions of file content.
• Smoother initialization: Existing folders are reused and missing ones are added; individual file issues are reported separately. Raw property or tag updates no longer interrupt analysis, and changed note bodies are read again. A model review that reads the sources and finds no new knowledge to add is also a valid result; unread sources remain pending.
• Clear progress and reasons: Folder preparation, file organization, and note analysis are listed with explicit actions, counts, and units. Pauses, interruptions, and specific errors are distinguished, with next steps matching the available buttons. Questions from the model can be answered on the initialization page.
• Yearly to-do heatmap: View total, today, and annual completion counts in To-dos settings. Switch years and hover over a day for its count. Older completions without a known date count toward the total only; no date is invented.
• Detail fixes: Improved settings layouts in smaller windows and consistent cards for folder actions. Fixed some avatar and identity save errors on Windows, custom configuration-folder paths, message measurement in pop-out windows, and feedback and state recovery after a failed favorite save.

Use and upgrade

Open Settings → EchoInk Agent → Knowledge → Folder management for Optimize names and Restore. Restore is unavailable without an original layout record. Annual completion history is in To-dos settings. Upgrading does not require vault reinitialization.

Check for updates in Obsidian's Community plugins page, or download main.js, manifest.json, and styles.css from this release, place them in the vault's .obsidian/plugins/codex-echoink/ folder, and re-enable the plugin. Keep your own data.json.

Requires Obsidian Desktop 1.11.4 or later. Mobile is not supported. Full real-app revalidation of the new folder-renaming and initialization-feedback flow remains pending, including corresponding Windows/Linux interaction. The main file still exceeds Obsidian Sync Standard's 5 MiB per-file limit; update each device separately if you use that plan. The cover is illustrative.

View on GitHub ↗

Ask when you need to. Pick up where you left off.

EchoInk Agent 2.3.0 adds a quick AI window and Markdown to-dos, with easier navigation for settings and long conversations.

Open a conversation beside your work

While Obsidian is running, use a global shortcut from another application to open the AI window and start typing. Move it wherever you need it. Closing the window returns the same conversation to the top tab group in Obsidian's right sidebar, keeping drafts, attachments, selected note references, and any response still being generated.

Enable it under Settings → EchoInk Agent → General → Quick AI window and choose a shortcut. The default is ⌘⇧E on macOS or Ctrl+Shift+E on Windows/Linux. Press the shortcut again or Esc to hide the window. Choose another combination if the shortcut is already occupied.

Keep to-dos in a note you can edit

The homepage now includes a to-do table with people, due dates, categories, and completion states. Combine category and completion filters; the compact progress bar shows completion for the current category.

Select Add to-do on the homepage. Manage categories or open the source file from the separate To-dos settings tab. Tasks live in EchoInk/待办.md in the current vault and can also be edited directly in Obsidian. Failed saves keep form input, and source-file edits made during table operations are preserved. Deleting a category moves its tasks to Uncategorized without deleting them.

Clearer navigation in a small window

Settings now use bookmark-style tabs: inactive tabs show an icon, while the selected tab rises and reveals its name before playing its existing icon animation. Hover for the full label. Layout & appearance has its own tab for homepage visibility controls. A new turn navigator along the right edge of long conversations lets you preview and jump to earlier turns.

Download and upgrade

• Check for updates in Obsidian's Community plugins page. For manual installation, download main.js, manifest.json, and styles.css from this release, place them in the vault's .obsidian/plugins/codex-echoink/ folder, and re-enable the plugin. Keep your own data.json.
• Upgrading from 2.2.2 does not require vault reinitialization. Conversations, Provider settings, and long-term memory are retained. Existing legacy to-dos are migrated to the Markdown source with a migration backup.
• Requires Obsidian Desktop 1.11.4 or later. Mobile is not supported. The quick window has been verified on macOS; real-device quick-window interaction on Windows/Linux remains unverified.
• The main file still exceeds Obsidian Sync Standard's 5 MiB per-file limit; update each device separately if you use that plan. The cover is an illustration; the plugin is the source of truth for UI and data.

View on GitHub ↗

Fixed an issue that prevented the plugin from being enabled on Windows.

A place to pick up where you left off.

An unfinished thought, a note you kept returning to this week, an idea you haven't sorted yet: EchoInk Agent 2.2.2 brings them together in a redesigned personal knowledge workspace. A complete settings refresh makes the rest of the plugin easier to navigate, while your familiar Obsidian folders, note tabs, and Agent sidebar stay in place.

Open your workspace and find your thread

Recent notes bring newly created and edited pages back into view. Search expands within the page, with suggestions appearing below as you type. For a passing idea, select Quick note to create and open a blank note in Inbox, ready to write.

Select a recent note or search result to open it in an Obsidian note tab. The workspace is a tab too: return whenever you need it, or reopen it from the left EchoInk entry if you close it.

See your progress as days you can return to

Weekly activity brings writing, editing, and rereading together. The calendar connects journals and activity to a date, giving you another way to recover a line of thought.

Select a day to see its records, or open View this week and follow an entry back to its note. Activity starts accumulating when tracking is enabled; matching actions on the same note and day are combined. Earlier activity is not invented or backfilled.

A home for your knowledge, a clear next step

Six cards—Wiki, Outputs, Projects, Inbox, Journal, and Review—connect knowledge, finished work, projects, ideas, daily writing, and reflection. Gentle colors distinguish the destinations, and a light hover animation reveals each action.

• Wiki / Outputs / Projects / Inbox: Open the native Obsidian local graph for that folder's index note and explore the notes connected to it.
• Journal: Choose a template to start writing. Prefer talking first? Use Journal with your Agent at the top of the workspace to reflect on your day with the Agent.
• Review: Start a fresh knowledge-review conversation in the right sidebar, revisiting notes and exploring connections one step at a time. Writing changes back to notes still requires your confirmation.

Settings with a consistent rhythm

All five settings tabs and their detail pages now share cards, spacing, buttons, switches, inputs, and selectors. Model connections, resources, knowledge status, and review tools each have a clear place. Layouts adapt to the available space as windows resize or the interface switches between Chinese and English.

Under API Provider, choose a service by group, enable models, and adjust defaults and parameters. Under Skills & MCP, search resources, open details, and edit configuration. Hover over a question mark beside a heading for an explanation when you need it.

Under Knowledge, inspect the local structure score, check history, and maintenance progress. The score uses actual directory, index-file, and source status. It describes structural readiness, not the quality of your writing or a comprehensive vault diagnosis.

Organize existing notes after seeing where they will go

Custom initialization gives each destination folder one row, so a large collection of notes no longer turns into a sprawling assignment table.

Open Knowledge → Custom plan, select the add action beside a folder, search by title or path, and choose multiple notes. Review assigned items, undo selections, and check the plan. Organization runs only after you select Start initialization, leaving room to adjust your choices first.

A clearer first meeting

The redesigned five-step guide uses focused highlights and progress cards that match the new workspace. It points to the actual left-side entry, settings, model connection, knowledge plan, and Agent profile.

Continue with Next, go back, jump between steps, or leave at any time. The completion screen lets you replay the guide. It shows you around; connecting a model and organizing notes remain your actions.

Keep working while another conversation finishes

While the EchoInk chat view remains open, switch to another conversation and continue working. The earlier conversation can keep running and shows an unread indicator when it finishes; runs can also be stopped individually. English now covers the workspace, settings, and chat sidebar, while your own notes and conversations keep their original language. The main plugin file has dropped from about 12.64 MiB in 2.1.0 to 8.71 MiB, a download-size reduction of roughly 31%.

Download and upgrade

• Upgrade directly from 2.1.0 without reinitializing Knowledge or rebuilding conversations and long-term memory. Existing Provider and model settings remain available.
• Check for updates in Obsidian's Community plugins page. For manual installation, download main.js, manifest.json, and styles.css from this release, place them in your vault's .obsidian/plugins/codex-echoink/ folder, and re-enable the plugin. Keep your existing data.json.
• Requires Obsidian Desktop 1.11.4 or later; mobile is not supported. The main file still exceeds Obsidian Sync Standard's 5 MiB per-file limit, so update each device separately when using that plan.
• The cover uses illustrative homepage content. Your actual notes, counts, activity, and status come from your own vault.

Full changes / 完整变更

View on GitHub ↗

A place to pick up where you left off.

An unfinished thought, a note you kept returning to this week, an idea you haven't sorted yet: EchoInk Agent 2.2.1 brings them together in a redesigned personal knowledge workspace. A complete settings refresh makes the rest of the plugin easier to navigate, while your familiar Obsidian folders, note tabs, and Agent sidebar stay in place.

Open your workspace and find your thread

Recent notes bring newly created and edited pages back into view. Search expands within the page, with suggestions appearing below as you type. For a passing idea, select Quick note to create and open a blank note in Inbox, ready to write.

Select a recent note or search result to open it in an Obsidian note tab. The workspace is a tab too: return whenever you need it, or reopen it from the left EchoInk entry if you close it.

See your progress as days you can return to

Weekly activity brings writing, editing, and rereading together. The calendar connects journals and activity to a date, giving you another way to recover a line of thought.

Select a day to see its records, or open View this week and follow an entry back to its note. Activity starts accumulating when tracking is enabled; matching actions on the same note and day are combined. Earlier activity is not invented or backfilled.

A home for your knowledge, a clear next step

Six cards—Wiki, Outputs, Projects, Inbox, Journal, and Review—connect knowledge, finished work, projects, ideas, daily writing, and reflection. Gentle colors distinguish the destinations, and a light hover animation reveals each action.

• Wiki / Outputs / Projects / Inbox: Open the native Obsidian local graph for that folder's index note and explore the notes connected to it.
• Journal: Choose a template to start writing. Prefer talking first? Use Journal with your Agent at the top of the workspace to reflect on your day with the Agent.
• Review: Start a fresh knowledge-review conversation in the right sidebar, revisiting notes and exploring connections one step at a time. Writing changes back to notes still requires your confirmation.

Settings with a consistent rhythm

All five settings tabs and their detail pages now share cards, spacing, buttons, switches, inputs, and selectors. Model connections, resources, knowledge status, and review tools each have a clear place. Layouts adapt to the available space as windows resize or the interface switches between Chinese and English.

Under API Provider, choose a service by group, enable models, and adjust defaults and parameters. Under Skills & MCP, search resources, open details, and edit configuration. Hover over a question mark beside a heading for an explanation when you need it.

Under Knowledge, inspect the local structure score, check history, and maintenance progress. The score uses actual directory, index-file, and source status. It describes structural readiness, not the quality of your writing or a comprehensive vault diagnosis.

Organize existing notes after seeing where they will go

Custom initialization gives each destination folder one row, so a large collection of notes no longer turns into a sprawling assignment table.

Open Knowledge → Custom plan, select the add action beside a folder, search by title or path, and choose multiple notes. Review assigned items, undo selections, and check the plan. Organization runs only after you select Start initialization, leaving room to adjust your choices first.

A clearer first meeting

The redesigned five-step guide uses focused highlights and progress cards that match the new workspace. It points to the actual left-side entry, settings, model connection, knowledge plan, and Agent profile.

Continue with Next, go back, jump between steps, or leave at any time. The completion screen lets you replay the guide. It shows you around; connecting a model and organizing notes remain your actions.

Keep working while another conversation finishes

While the EchoInk chat view remains open, switch to another conversation and continue working. The earlier conversation can keep running and shows an unread indicator when it finishes; runs can also be stopped individually. English now covers the workspace, settings, and chat sidebar, while your own notes and conversations keep their original language. The main plugin file has dropped from about 12.64 MiB in 2.1.0 to 8.71 MiB, a download-size reduction of roughly 31%.

Download and upgrade

• Upgrade directly from 2.1.0 without reinitializing Knowledge or rebuilding conversations and long-term memory. Existing Provider and model settings remain available.
• Check for updates in Obsidian's Community plugins page. For manual installation, download main.js, manifest.json, and styles.css from this release, place them in your vault's .obsidian/plugins/codex-echoink/ folder, and re-enable the plugin. Keep your existing data.json.
• Requires Obsidian Desktop 1.11.4 or later; mobile is not supported. The main file still exceeds Obsidian Sync Standard's 5 MiB per-file limit, so update each device separately when using that plan.
• The cover uses illustrative homepage content. Your actual notes, counts, activity, and status come from your own vault.

Full changes / 完整变更

View on GitHub ↗

A place to pick up where you left off.

An unfinished thought, a note you kept returning to this week, an idea you haven't sorted yet: EchoInk Agent 2.2.0 brings them together in a redesigned personal knowledge workspace. A complete settings refresh makes the rest of the plugin easier to navigate, while your familiar Obsidian folders, note tabs, and Agent sidebar stay in place.

Open your workspace and find your thread

Recent notes bring newly created and edited pages back into view. Search expands within the page, with suggestions appearing below as you type. For a passing idea, select Quick note to create and open a blank note in Inbox, ready to write.

Select a recent note or search result to open it in an Obsidian note tab. The workspace is a tab too: return whenever you need it, or reopen it from the left EchoInk entry if you close it.

See your progress as days you can return to

Weekly activity brings writing, editing, and rereading together. The calendar connects journals and activity to a date, giving you another way to recover a line of thought.

Select a day to see its records, or open View this week and follow an entry back to its note. Activity starts accumulating when tracking is enabled; matching actions on the same note and day are combined. Earlier activity is not invented or backfilled.

A home for your knowledge, a clear next step

Six cards—Wiki, Outputs, Projects, Inbox, Journal, and Review—connect knowledge, finished work, projects, ideas, daily writing, and reflection. Gentle colors distinguish the destinations, and a light hover animation reveals each action.

• Wiki / Outputs / Projects / Inbox: Open the native Obsidian local graph for that folder's index note and explore the notes connected to it.
• Journal: Choose a template to start writing. Prefer talking first? Use Journal with your Agent at the top of the workspace to reflect on your day with the Agent.
• Review: Start a fresh knowledge-review conversation in the right sidebar, revisiting notes and exploring connections one step at a time. Writing changes back to notes still requires your confirmation.

Settings with a consistent rhythm

All five settings tabs and their detail pages now share cards, spacing, buttons, switches, inputs, and selectors. Model connections, resources, knowledge status, and review tools each have a clear place. Layouts adapt to the available space as windows resize or the interface switches between Chinese and English.

Under API Provider, choose a service by group, enable models, and adjust defaults and parameters. Under Skills & MCP, search resources, open details, and edit configuration. Hover over a question mark beside a heading for an explanation when you need it.

Under Knowledge, inspect the local structure score, check history, and maintenance progress. The score uses actual directory, index-file, and source status. It describes structural readiness, not the quality of your writing or a comprehensive vault diagnosis.

Organize existing notes after seeing where they will go

Custom initialization gives each destination folder one row, so a large collection of notes no longer turns into a sprawling assignment table.

Open Knowledge → Custom plan, select the add action beside a folder, search by title or path, and choose multiple notes. Review assigned items, undo selections, and check the plan. Organization runs only after you select Start initialization, leaving room to adjust your choices first.

A clearer first meeting

The redesigned five-step guide uses focused highlights and progress cards that match the new workspace. It points to the actual left-side entry, settings, model connection, knowledge plan, and Agent profile.

Continue with Next, go back, jump between steps, or leave at any time. The completion screen lets you replay the guide. It shows you around; connecting a model and organizing notes remain your actions.

Keep working while another conversation finishes

While the EchoInk chat view remains open, switch to another conversation and continue working. The earlier conversation can keep running and shows an unread indicator when it finishes; runs can also be stopped individually. English now covers the workspace, settings, and chat sidebar, while your own notes and conversations keep their original language. The main plugin file has dropped from about 12.64 MiB in 2.1.0 to 8.71 MiB, a download-size reduction of roughly 31%.

Download and upgrade

• Upgrade directly from 2.1.0 without reinitializing Knowledge or rebuilding conversations and long-term memory. Existing Provider and model settings remain available.
• Check for updates in Obsidian's Community plugins page. For manual installation, download main.js, manifest.json, and styles.css from this release, place them in your vault's .obsidian/plugins/codex-echoink/ folder, and re-enable the plugin. Keep your existing data.json.
• Requires Obsidian Desktop 1.11.4 or later; mobile is not supported. The main file still exceeds Obsidian Sync Standard's 5 MiB per-file limit, so update each device separately when using that plan.
• The cover uses illustrative homepage content. Your actual notes, counts, activity, and status come from your own vault.

Full changes / 完整变更

View on GitHub ↗

A personal Agent with an identity that grows over time

1. Choose from eight starting styles, 15 built-in avatars, or a custom SVG avatar. The updated guide lets you name the Agent and choose its avatar; changing either later does not clear long-term memory or learned habits.
2. Long-term memory can now add and update useful records during ordinary conversations without requiring a repeated “remember this” instruction. Scheduled offline memory organization connects related experiences and gradually updates how the Agent understands and works with you.
3. Open Settings → EchoInk Agent → General → Long-term memory / Identity and user profile to view Agent and user profiles, choose another starting style, and set offline organization from one to six times per day. Turning off offline organization leaves normal memory writes and recall available. Turning off long-term memory keeps the Agent's name, avatar, and base style.

Images, documents, and vault notes in conversation

1. Paste or drag images into the composer, or use + to add images and files. Images are sent only when the current model supports them; BMP, HEIC, HEIF, and SVG are converted to PNG when possible.
2. Attach PDF, Word, Markdown, and HTML documents. A turn accepts up to eight files, 20 MiB per file and 50 MiB in total. Encrypted, damaged, or over-context documents are rejected with a clear message before sending. OCR for scanned PDFs is not included yet.
3. Type @ to find Markdown notes in the current vault by file name, path, alias, Pinyin, or initials and add a selected note to the current turn. Sent images and documents remain available as openable thumbnails or file cards.

A clearer conversation process

1. Expand live reasoning while the model works. Tool progress, approval requests, task plans, file changes, and the final answer share one timeline. Manually scrolling up no longer forces the reasoning view back to the bottom.
2. For complex work, open Plan mode from the composer's + menu. Approve or reject requested tool access directly inside the conversation.
3. Knowledge, Memory, note, and document sources are shown with the turn so you can see which background informed the answer.

Providers and models

1. Added OpenAI Codex Beta browser sign-in. Open Settings → EchoInk Agent → API Provider → Add API Provider → Sign in, choose OpenAI Codex Beta, and select Sign in with OpenAI. If the browser cannot return automatically, paste the callback URL or authorization code to finish.
2. Qwen API and Qwen Token Plan now have separate, clearly named entries, with fixes for Token Plan transport and stream completion.
3. Save multiple instances of the same Provider, enable several models per instance, choose a default, discover available models, or add a Model ID manually. Deep-reasoning controls appear only when the selected model supports them.

Setup and Knowledge initialization

1. A one-time five-step guide now covers opening the Agent sidebar, entering settings, connecting a model, setting up Knowledge, and choosing an Agent style. Dismiss it at any time or restart it from settings.
2. Knowledge setup shows a preview before organizing anything. Missing standard folders can be restored without moving, deleting, or rewriting existing notes, and a visual guide is available in settings.

Reliability fixes

1. Fixed cases where some Providers could keep reasoning or repeat an answer after a short follow-up, especially when Memory, Knowledge, notes, or documents were included. Current questions and relevant background are now ordered consistently, and a turn can finish normally even when no Knowledge reference is found.
2. Failed or interrupted model output no longer leaks into later conversation history. Tool completion, clean stream endings, long-conversation recovery, consecutive memory writes, and reasoning-panel scrolling are also more reliable.

Upgrade notes

• Upgrade directly from 2.0.3 to 2.1.0 without rebuilding conversations, Knowledge, or long-term memory. Saved Provider and model settings are upgraded automatically.
• 2.1.0 does not rewrite content already stored in old conversations. If an older conversation already contains repeated answers or failed turns, create a new conversation after upgrading.
• main.js in 2.1.0 is about 12.64 MiB, above the 5 MiB per-file limit of Obsidian Sync Standard. Local installation and Community downloads still work, but update the plugin separately on each device or use a sync method that supports larger files.
• Obsidian Desktop 1.11.4 or later is still required. Mobile is not supported.

View on GitHub ↗

1. Fixed the heading and static-style errors flagged by the Obsidian Community review: the settings page and section headings now use the Obsidian Setting API setHeading(), and the modal preformatted body styles moved into styles.css.
2. Strengthened the local review gate: every obsidianmd/* error now fails hard instead of being allowed by the baseline.
3. Fixed OpenAI- and Anthropic-compatible Providers failing before requests were sent, restoring cloud conversations.
4. Known limitation: main.js is about 5.71 MiB, above the 5 MiB per-file limit of Obsidian Sync Standard. Local installation and Community downloads still work, but Standard Sync cannot synchronize this file.

View on GitHub ↗

1. Significantly reduced the plugin bundle (main.js) to restore compatibility with the Obsidian Sync Standard plan.
2. Removed the unused Pi CLI, tool download installer, and self-update code that EchoInk does not use.
3. Fixed the community directory Manifest description by removing a redundant word.

View on GitHub ↗

EchoInk Agent

1. The public plugin name is now EchoInk Agent; Codex EchoInk is no longer used as the product name.
2. The public description now presents EchoInk Agent as a personal knowledge agent that organizes and maintains Obsidian knowledge, builds visible and correctable long-term memory, and gradually learns what matters to the user.
3. The plugin ID and GitHub repository remain unchanged, so existing users can upgrade in place without migrating conversations, knowledge, or memory.

Direct Provider setup

1. Normal installation does not require Codex CLI. Add the API URL, API key, and model for OpenAI or a compatible service in Provider settings to get started.
2. Conversations, knowledge maintenance, long-term memory, memory correction, and review continue to work as before.

View on GitHub ↗

One conversation and Agent path

1. Conversations now use one Pi-native AgentSession path. Opening or switching a conversation restores local history while activating its AgentSession, so it is ready to continue.
2. Conversations can be created, renamed, archived, restored, and soft-deleted. Archived conversations moved from the chat dialog to Settings → EchoInk → Review, while the original Pi Session JSONL remains intact.
3. The conversation header, message actions, and composer toolbar were rebuilt. The send button is smaller, and low-value Agent status and knowledge-health widgets were removed.

Visible, correctable long-term memory

1. Current Memory is grouped into Facts, Views, Decisions, Active, and Episodes. Each record shows its title, content, and recall context without exposing internal paths or revisions.
2. The correction dialog keeps the original memory, correction input, and cyan corrected preview visible together. Preview generation can be stopped and never writes the original record.
3. Only Save creates a new version. Editing the input immediately invalidates an old preview; closing, stopping, failure, late results, and revision conflicts cannot overwrite current Memory.

Direct knowledge maintenance and review

1. /maintain without arguments runs globally. Selecting one Raw note with the composer's + targets that note, while trailing text fuzzy-searches related Raw notes.
2. Already-refined content with no changes settles without another write. Candidates are bound to target revisions, preventing a generated result from overwriting newer content.
3. Report generation and destination settings now share one Review page. Choose Agent or knowledge-base reports, the date range, any vault folder, and whether to open the generated HTML. The recent-report list has been removed.

Providers and upgrade notes

1. Provider setup now stores API keys directly and no longer requires a separate Credential step. Existing configurations that contain only a retired Credential reference require the API key to be entered once.
2. EchoInk 2.0 does not read or migrate retired Codex, OpenCode, or Hermes conversations, or retired Cognitive, Reflection, and Memory data. It does not proactively delete those old files.
3. Obsidian 1.11.4 or later is required, and EchoInk remains desktop-only. Keep a vault snapshot before upgrading, following your normal backup practice.

View on GitHub ↗

More durable conversations and safer Knowledge maintenance: v1.4.0 keeps Chat and Knowledge conversations on one recoverable history across storage migration, context changes, plugin reloads, and Agent switches. Knowledge maintenance now isolates and validates output before it reaches the Vault.

Conversation lifecycle upgrade

• Existing Chat and Knowledge conversations migrate automatically while preserving messages, workspace boundaries, and recoverable Agent bindings.
• New blank chats, existing Chat sessions, and direct Knowledge conversations can all be saved and continued after a plugin reload.
• Conversation threads, individual turns, and native Agent sessions use distinct identities, reducing recovery conflicts and cross-session mix-ups.

Safer Knowledge maintenance

• /maintain starts with the selected Agent and tries a fallback only when the failed attempt is safely isolated and contains no committable result.
• Maintenance output is validated in isolation before a recoverable commit reaches the Vault. Timeouts, reloads, and partial failures do not commit the same result twice.
• Oversized sources, unchanged inputs, and recovering tasks now receive accurate states instead of being marked as processed or reduced to generic failures.
• Maintenance progress and final reports are easier to follow. Reports expand and collapse reliably, and referenced Vault notes open directly for verification.
• Fixed execution diagnostics being mistaken for a divergence between the live conversation and its durable record. Scheduled maintenance also stops appending background messages to active conversations.
• Only explicitly retryable scheduled failures run again, with at most three daily attempts after 5- and 15-minute delays instead of retrying every minute.
• EchoInk disables global hooks and global memory injection for the Codex processes it launches, preventing external memory layers from taking over maintenance Shadows. EchoInk's own .echoink/memory remains unchanged.
• Fixed recovery being blocked when Obsidian updated the updated field during a maintenance transaction and the expected install hardlink was mistaken for external tampering. A genuine additional hardlink still blocks the commit.

Record and session management

• Conversation history, workflow runs, business artifacts, and native Agent sessions follow separate retention, recovery, and cleanup rules.
• A compact session picker adds search, keyboard navigation, and batch management while protecting the Knowledge channel and running sessions.
• Cleanup operations produce a recoverable plan and evidence before applying changes, reducing accidental deletion and half-finished states.

How to use

1. Install v1.4.0 and reload Obsidian. Existing conversations upgrade automatically when needed; no manual migration is required.
2. Open a new or existing conversation. Select a workspace for regular Chat when prompted; the Knowledge channel continues to use the current Vault.
3. Run /check or /maintain in Knowledge, follow the stage progress, then expand the report and open referenced notes to verify the result.
4. Use the session entry in the header to search or manage existing conversations. Running sessions and the Knowledge channel are protected from batch deletion.

Existing Vault notes and EchoInk conversations do not require a reset. Legacy Native Execution cleanup warnings may still appear for a small amount of historical data, but they do not change saved conversations or business results.

Download

Obsidian reads these single-file Release assets directly: main.js, manifest.json, and styles.css.

View on GitHub ↗

A production-ready multi-Agent workspace: v1.3.0 brings setup, runtime status, conversation projection, local memory, and utility-model configuration for Codex, OpenCode, and Hermes into one EchoInk experience. Switching backends changes the execution engine without changing the conversation layout.

Three backends, one runtime

• Added a unified setup and connection dashboard for Codex, OpenCode, and Hermes, with direct install, repair, and recheck paths for missing requirements.
• All three backends now share the same presentation for public reasoning, tool calls, processing time, and final answers. Keep the process collapsed for answer-first reading or expand it for full detail.
• Hardened OpenCode session recovery, Hermes tool terminal states, and interrupted runs so a dropped stream does not execute the same task twice.

Memory V2 and Knowledge rules

• Added EchoInk Memory V2 for locally curated, cross-session and cross-backend recall while preserving each Agent's native memory.
• Every Knowledge run now reloads the configured Markdown rules file. Missing, empty, unreadable, out-of-vault, or invalid files stop before the Agent starts.
• Ordinary Vaults no longer depend on AGENTS.md. LLM-WIKI.md remains the default, and another Vault Markdown file can be selected in settings.

Prompt enhancement and utility models

• Prompt enhancement now has its own Agent backend and model settings without changing the main chat model.
• The enhancer model is now selected from a list. Use Add model to enter a model ID, add it to the current backend, and select it immediately.
• Codex defaults to gpt-5.6-terra, medium reasoning, and fast responses. OpenCode and Hermes show only the model, Agent, or Profile controls they actually support.

Reliability fixes

• Fixed interrupted Memory transactions leaving incomplete state, and clarified the boundary between local history and native Agent memory.
• Fixed incompatible model, Agent, or Profile selections carrying across prompt-enhancer backends.
• Fixed unsupported reasoning or fast-response controls appearing for backends that do not provide them.

How to use

1. Install v1.3.0 and reload Obsidian.
2. Open EchoInk settings and confirm that at least one Agent is ready. Install or recheck additional backends when needed.
3. Switch between Codex, OpenCode, and Hermes from the sidebar header without clearing the current EchoInk session.
4. Keep the automatic prompt-enhancer model, or use Add model to register your own model ID.
5. Initialize and sync Memory V2 in settings when you want local recall across sessions.

Existing Vault files and EchoInk sessions do not require migration.

Download

Obsidian reads these single-file Release assets directly: main.js, manifest.json, and styles.css.

View on GitHub ↗

Review compatibility fix: v1.2.2 resolves the Obsidian automated source-review failure caused by Agent parameter-menu styling. Menu positioning, visibility, and existing interactions remain unchanged.

What's New

• Updated parameter menus to use Obsidian-approved styling APIs.
• Preserved menu positioning, visibility, and existing interactions.
• Preserved the lightweight Knowledge command menu and keyboard navigation introduced in v1.2.1.

How to Use

1. Install v1.2.2 and reload Obsidian.
2. Existing Vault files, sessions, and plugin settings do not require migration.

Download

Obsidian reads these single-file Release assets directly: main.js, manifest.json, and styles.css.

View on GitHub ↗

Lightweight Knowledge command menu: Typing / in the Knowledge channel now opens a cleaner Codex-style list: transparent by default, neutral in color, and highlighted only for the active command.

The screenshot above shows the updated menu in a real Obsidian workspace.

What's New

• Replaced heavy background cards and pink command text with black command names, neutral-gray icons, and concise descriptions.
• Added wraparound ArrowUp / ArrowDown navigation with automatic scrolling for long lists.
• Enter fills the selected command without sending it; Escape closes the menu to reduce accidental runs.

How to Use

1. Install v1.2.1 and reload Obsidian.
2. Open the Knowledge channel and type / in the composer.
3. Select a command with ArrowUp / ArrowDown, then press Enter to fill it before adding parameters or sending.

Download

Obsidian reads these single-file Release assets directly: main.js, manifest.json, and styles.css.

View on GitHub ↗

Unified Agent Harness and rebuilt UI: Codex, OpenCode, and Hermes now share one EchoInk runtime contract, context path, and conversation layout. The sidebar has been rebuilt around answer-first reading, with independent prompt enhancement, fast utility-model routing, and one Bookmark entry.

The real Obsidian screenshots above are placed beside the changes they demonstrate: the full Home workspace and Agent sidebar, the Codex / OpenCode / Hermes switcher, and the Plan-mode composer.

What's New

• Routed Codex, OpenCode, and Hermes through one EchoInk Harness without clearing the current session when the main Agent changes.
• Made final answers the primary reading path while reasoning, commands, file edits, and MCP calls stay in an expandable process timeline.
• Added a three-Agent header switcher and a lighter responsive composer for workspace, Plan state, model, reasoning, and speed.
• Gave prompt enhancement its own Agent, provider, API path, and model settings; rewrite, expand, and continue default to fast utility models.
• Combined WeChat article and public web-page capture into one Bookmark action with automatic URL routing.
• Fixed history scroll snapping, Knowledge card jitter, unclear timeout states, and narrow-sidebar popover overflow.

How to Use

1. Install v1.2.0 and reload Obsidian.
2. Choose Codex, OpenCode, or Hermes from the EchoInk header.
3. Use Bookmark, Skill, or prompt enhancement from the lower-left composer actions; change model, reasoning, speed, and Agent / Plan mode from the lower-right menu.
4. Open plugin settings only when you want to override the automatic fast model used by prompt enhancement or writing actions.

Download

Obsidian reads these single-file Release assets directly: main.js, manifest.json, and styles.css.

View on GitHub ↗

Agent tools and Knowledge digest update: EchoInk now makes Agent activity easier to follow, connects the current vault's tool resources more clearly, and guides Knowledge maintenance through a stricter four-step digest flow.

What's New

• Added a tool broker foundation for vault resources, MCP tools, Skills, and tool bundles, with clearer per-scope switches.
• Improved the Agent process timeline so searches, file work, tool calls, and completion states are easier to understand.
• Added a stricter Knowledge digest path: understand Raw, extract reusable knowledge, merge it into Wiki / Projects, then mark Raw only after source evidence is checked.
• Added the first Hermes backend entry point, while keeping Hermes model and provider setup in Hermes itself.
• Split the large Agent sidebar view into smaller UI modules, making future UI review and maintenance safer.

How to Use

1. Install v1.1.0.
2. Open EchoInk in Obsidian and choose the Agent backend you already use.
3. Use /check, /maintain, or /reingest in the Knowledge channel when you want EchoInk to inspect or digest Raw notes.
4. Open settings to review resource switches for chat, Knowledge, and writing actions.

Download

Obsidian reads these single-file Release assets directly: main.js, manifest.json, and styles.css.

View on GitHub ↗

Review style fix: This update addresses the Obsidian community review rule for direct style assignment without changing the EchoInk workflow.

What's New

• Replaced direct sidebar style assignments with Obsidian-supported setCssStyles and setCssProps.
• Kept the health tooltip, yearly heatmap, virtual message list, and context usage ring behavior unchanged.
• Added source scanning coverage to prevent direct style assignment from returning to the Agent sidebar.
• The Release provides only the official assets main.js, manifest.json, and styles.css, with GitHub artifact attestations.

How to Use

1. Install v1.0.3.
2. Open EchoInk Home or the Codex sidebar as usual.

Download

Obsidian reads these single-file Release assets directly: main.js, manifest.json, and styles.css.

View on GitHub ↗

Review compatibility release: This update addresses Obsidian community review findings around view lifecycle and compatibility without changing the core EchoInk workflow.

What's New

• Adjusted Obsidian view registration to avoid caching view instances on the plugin object and reduce lifecycle misuse risk.
• Plugin unload no longer force-closes EchoInk panes, so reloading the plugin is less likely to disrupt the workspace layout.
• Removed dependency on newer Obsidian APIs and kept compatibility with the declared minAppVersion.
• The Release provides only the official assets main.js, manifest.json, and styles.css, with GitHub artifact attestations.

How to Use

1. Install v1.0.2 and open EchoInk Home, the Codex sidebar, or review preview as usual.
2. If you previously saw a community review failure, wait for Obsidian to re-check the latest Release.

Download

Obsidian reads these single-file Release assets directly: main.js, manifest.json, and styles.css.

View on GitHub ↗

Small fixes release: This update improves Home calendar behavior, Knowledge conflict checks, and large-file read boundaries to reduce accidental processing and Obsidian slowdowns.

What's New

• Home calendar now supports previous month, next month, and return-to-current-month actions.
• Knowledge maintenance pre-checks duplicate Wiki pages with numeric suffixes such as Title 2.md and moves conflicts into outputs/maintenance/conflict-duplicates-*.
• Maintenance prompts now explicitly prevent numeric suffixes from bypassing an existing Wiki page; matching pages must be updated or reported as conflicts.
• Dashboard, Raw discovery, and /ask now use stricter file-read budgets so large PDFs, images, and very large Markdown files are not fully loaded.
• Raw files that exceed the read budget are left out of the current source set and are not written to the tracker or marked as processed.

How to Use

1. After installing v1.0.1, run /maintain to let EchoInk pre-check existing numeric-suffix Wiki conflicts.
2. Review the maintenance report for moved conflict files under outputs/maintenance/conflict-duplicates-*.
3. If a large Raw file is skipped, split it or process it separately in a later run.

Download

Obsidian reads these single-file Release assets directly: main.js, manifest.json, and styles.css.

View on GitHub ↗

Home dashboard update: EchoInk adds a dedicated Home dashboard so you can see your Obsidian knowledge base status, pending Raw notes, activity, and recent knowledge cards in one place.

What's New

• Added an EchoInk Home tab that can open by default, be closed, and be reopened from the Obsidian command palette.
• The ribbon entry now opens both the EchoInk sidebar and the Home dashboard.
• Home shows Wiki status, Raw pending work, yearly check heatmap, calendar, health score, and key vault counts.
• Added a responsive note-card stream that prioritizes recent Wiki updates and falls back to recommendations.
• Cards can be filtered by status, recommendation group, update time, relevance, and first-level folder.
• Each card menu can copy Obsidian internal links, relative paths, and Markdown links.

How to Use

1. Open Obsidian and use Home as the knowledge-base workspace.
2. Use the filters and sorting controls above the cards to focus on recent updates, pending work, or a specific folder.
3. Use each card menu to copy the link format you need.

Download

Obsidian reads these single-file Release assets directly: main.js, manifest.json, and styles.css.

View on GitHub ↗

Codex EchoInk 0.8.0

Knowledge safety update: Raw sources, local history, and background Codex threads are now handled as separate layers.

New in this release

• Raw sources are protected more strictly during /check, /maintain, and /calibrate raw.
• Knowledge history stays readable from EchoInk's local history store through /history.
• Background Codex threads created by Knowledge commands are archived after task results are saved, reducing recent-list clutter in Codex Desktop.
• Canceled, timed-out, retried, or failed Knowledge runs now leave clearer state in the UI.
• The Knowledge dashboard Raw count now focuses on actual source notes instead of .assets/ image files.

How to use

1. Run /check for a read-only Knowledge health check.
2. Run /maintain to process changed Raw sources.
3. Use /history to browse local Knowledge history. Removing Codex archived conversations does not remove plugin history.

<details> <summary>Download verification</summary>

• codex-echoink-0.8.0.zip
• SHA256: 2f4924b8db41b3daf5f45439741e6923c952086d268e691f66197af1d06096e9

</details>

View on GitHub ↗

v0.7.2

Turn Queue and composer menu stability update for Codex EchoInk.

What changed

• Clicking inside the composer area but outside Skill and Knowledge command menus now closes open composer menus.
• Clicking a Skill or Knowledge command menu no longer closes that menu by mistake.
• The queue entry UI has regression coverage so follow-up task entry stays stable.

How to use

1. Open the Codex EchoInk sidebar.
2. Click the Skill button or type a Knowledge command trigger.
3. Use the menu normally; click back into the composer when you want it to close.

<details> <summary>Download check</summary>

• Asset: codex-echoink-0.7.2.zip
• SHA256: 3e55edc7b31f0b5ae37e82be3ce7bf5fd23602a67d5af0046b1b79ec56be200b

</details>

View on GitHub ↗

v0.7.1

Turn Queue stability update for Codex EchoInk.

What changed

• Successful tasks continue the queue only when another item is waiting.
• Failed or stopped tasks pause the queue and keep remaining work for manual resume.
• Queued turns no longer start while an ordinary turn, Knowledge task, or queue startup is already in progress.
• Dragging queue cards stays inside the queue UI instead of leaking into the composer attachment drop area.

How to use

1. Start a chat or Knowledge command.
2. Type the next task while the current one is running.
3. Click Enqueue.
4. Reorder or delete waiting items above the composer.
5. If a task is stopped or fails, click Resume queue when ready.

<details> <summary>Download check</summary>

• Asset: codex-echoink-0.7.1.zip
• SHA256: 65639acbd516b42632417fa3683e36fbc12f70f5802a2b9af91302ad6fc20d91

</details>

View on GitHub ↗

Codex EchoInk 0.7.0

Turn Queue for ordinary chat and Knowledge channel work.

What changed

• Queue follow-up tasks while the current Agent turn is still running.
• The primary button becomes Enqueue when a task is running and the composer has text.
• With an empty composer, the primary button still stops the current task.
• Queue items preserve their text, attachments, Skill, model, permission, mode, and workspace.
• Successful tasks automatically run the next queued item.
• Failed or stopped tasks pause the queue and keep the remaining items for manual resume.
• Knowledge commands such as /ask, /maintain, and /journal now run serially.

How to use

1. Start an ordinary chat or Knowledge command.
2. Type the next task while the current one is running.
3. Click Enqueue.
4. Reorder or delete waiting items above the composer.
5. If a task stops or fails, click Resume queue.

<details> <summary>Download verification</summary>

• Asset: codex-echoink-0.7.0.zip
• SHA256: a8b39d62b6c8f6c3d4b6b74ce4e6bbbfa61da9f782eba67ada5e4b0a1f8e94b2

</details>

View on GitHub ↗

Version

0.6.0

New features

• Added a first-run setup guide that checks Codex CLI, Codex login, OpenCode CLI, OpenCode server, models, and Agent readiness.
• Added missing-environment guidance with install commands, copy buttons, and documentation links.
• Added Run check again to re-detect CLI paths, refresh Codex login, and reconnect or start OpenCode when needed.
• Added Start as an explicit setup completion step. It opens the EchoInk sidebar without sending a message or running a Knowledge task.

Fixes and maintenance

• Added Windows path detection for Codex CLI and OpenCode CLI.
• Upgraded Knowledge history to day-based archive storage with settings tools for indexing, export, and compaction.
• Tightened Knowledge maintenance so Agent tasks cannot directly rewrite Raw source bodies; raw path normalization is handled by plugin-side checks.
• Improved Knowledge maintenance reports, dashboard state, local note links, and history entry placement.

How to use

1. Download codex-echoink-0.6.0.zip.
2. Unzip it into <vault>/.obsidian/plugins/codex-echoink/.
3. Restart Obsidian and enable Codex EchoInk.
4. Open plugin settings, follow the setup guide, then click Start.

<details> <summary>Download checksums</summary>

62cfc5dc9be8f292f20636f1770892ebfccaa31cf3e33be9cb6fd4d685de3d59  codex-echoink-0.6.0.zip
ede96fa6af225222ddad283715e0443899755da41fc014c65a9d6bb6f531fe70  main.js
df5f59d69ae9e4b1a607181b3da7e558fec842c61e9d93774070014632507991  manifest.json
f0e9d8e49ffa83be698ee1ee83fc8e8d314fcea29445853680b6cc64acce1406  styles.css
31e6edb9dd8c9a08318a9a0f9ad7647937dad989271435eeb7d626f299f39ba7  codex-echoink-v0.6.0-card.svg

</details>

View on GitHub ↗

v0.5.2

Knowledge workflow and diagnostics update.

Same-version update

• In the Knowledge channel, ordinary messages now stay as normal Agent chat.
• Only explicit /ask, /query, /问, or /查询 commands trigger Knowledge Q&A.
• The Knowledge channel primary button now stops ordinary Agent chat when that chat is running, instead of canceling Knowledge maintenance by mistake.
• Knowledge failures now keep more complete app-server, JSON-RPC, OpenCode, and turn error details.
• Local vault note paths and report paths in Knowledge replies render as clickable note links.

New Features

• Added Knowledge weekly reviews and Agent chat weekly reviews.
• Weekly reviews can run on schedule or manually, and generate Markdown plus HTML in outputs/obsidian-weekly-review/.
• Added EchoInk's built-in HTML preview for generated review reports.
• Added /week and /week agent shortcuts in the Knowledge channel.
• Upgraded /journal to write into the current journal/daily/YYYY-MM/YYYY-MM-DD-周X.md layout.
• /journal now creates missing journal folders and uses a fixed 00:00 to next-day 06:00 work window.
• OpenCode API mode can collect OpenCode chat history for /journal.
• Expanded /ask evidence from wiki/ to wiki/, journal/, and outputs/.
• Added citation buckets, excerpt lines, relevance labels, and match reasons for local Knowledge answers.
• Added a Chinese / English settings language switch.

Fixes

• Changed the Codex CLI default model to Auto.
• Existing saved gpt-5.5 defaults are migrated to Auto.
• Removed remaining hard-coded gpt-5.5 fallback paths from Knowledge tasks and Plan mode.
• Added detailed diagnostics for WebSocket failures, proxy refusal, missing CLI, timeouts, and app-server exits.
• Added troubleshooting guidance for Windows responses_websocket / os error 10061.
• Simplified Review settings so manual report generation has confirmation and clearer output paths.
• Codex CLI Knowledge runs now use the normal process timeline, so reasoning, commands, file edits, and final results are easier to inspect.

Install

Download codex-echoink-0.5.2.zip, unzip it, and place the codex-echoink folder in:

<vault>/.obsidian/plugins/codex-echoink/

Assets

• codex-echoink-0.5.2.zip
• main.js
• manifest.json
• styles.css

SHA256:

7fcda68dd2f4683b7984379956231b0883e038b6e239df48f6137c77a7fd7205  codex-echoink-0.5.2.zip
19953e426224c6b810f039001aa509c7aaa1dd098bf4aabb1d4f8f0c491f99f4  main.js
f06f93be57fed45fc874f583f143c20d9593e42f004018b3f314ad6fee8c043d  manifest.json
5c896e5f1f6a9e0d8237adc5b41655fe64c4507ea8b69f8d8c149b559b919f0c  styles.css

View on GitHub ↗

Community review fix.\n\n- Removed the redundant word "Obsidian" from manifest.json description to satisfy automated community checks.\n- Rebuilt release assets for version 0.5.1.\n\nAssets:\n- main.js\n- manifest.json\n- styles.css\n- codex-echoink-0.5.1.zip

View on GitHub ↗

Version

0.5.0

New features

• Renamed the plugin to Codex EchoInk and prepared the community plugin id codex-echoink.
• Added clearer privacy and permission disclosures for Codex CLI, OpenCode, model providers, API keys, and vault write boundaries.
• Kept compatibility with large raw message files from older manual installs under .obsidian/plugins/obsidian-codex/raw.
• Added community-ready release assets: main.js, manifest.json, styles.css, and codex-echoink-0.5.0.zip.

How to use

1. Install Codex CLI and log in for Codex CLI mode.
2. Optionally install OpenCode if you want OpenCode API mode for knowledge base tasks.
3. In Obsidian, enable Codex EchoInk and open the EchoInk sidebar.
4. Open the Knowledge channel and start with /init, /check, or /ask.

View on GitHub ↗

v0.4.1 - 知识库频道增强

这版重点增强 知识库 常驻频道，让 Obsidian vault 的问答、体检、维护和能力开关更容易日常使用。

新增功能

• 新增 /ask 只读问答：先检索 wiki/ 相关笔记，再回答问题，并区分 Vault 依据和补充信息。
• 支持自然问题自动进入问答流程，不必强记命令。
• 知识库健康热力图升级为 GitHub 风格年度视图，展示月份、星期、成功和失败状态。
• Codex CLI 模式下，知识库频道底部可直接选择模型和思考强度。
• 当前 vault 的 插件、MCP、Skills 能力管理支持搜索。
• 长路径和长描述会自动省略，右侧勾选框保持可见可点。
• 默认知识库规则文件切到 LLM-WIKI.md；AGENTS.md 保留为兼容选项和运行层背景。

使用方法

1. 打开 Codex 侧栏里的 知识库 频道。
2. 输入 /ask 你的问题，或直接在知识库频道里问问题。
3. 使用 Codex CLI 模式时，在输入框右下角选择模型和思考强度。
4. 展开知识库健康仪表盘，查看年度体检热力图。
5. 到插件设置的当前 vault 能力管理里，搜索 插件、MCP 或 Skills 后再勾选。

下载校验

• 安装包：obsidian-codex-0.4.1.zip
• SHA256：d20dc098893bc02128703e336a34b2b88b36830e3ef956c3cbb592e76f7c08fa

View on GitHub ↗

v0.4.0

This release adds the first Knowledge Base Operations workflow for maintaining an Obsidian vault from inside the Codex sidebar.

New

• Persistent Knowledge channel bound to the current Obsidian vault.
• Chat-first command templates: /check, /maintain, /outputs, /journal, /inbox.
• WeChat, web page, and file capture entry points for Raw Sources.
• Configurable knowledge base rules file: default AGENTS.md, or a custom Markdown file such as CLAUDE.md.
• OpenCode API mode model selection.
• OpenCode Agent selection.
• Editor action: translate selected text to English.

How To Use

1. Download and unzip obsidian-codex-0.4.0.zip.
2. Move the obsidian-codex folder into <vault>/.obsidian/plugins/.
3. Restart Obsidian and enable Codex for Obsidian.
4. Open the Knowledge channel in the Codex sidebar.
5. Use /check for a safe health check, /maintain to process new Raw Sources, and /outputs to turn outputs into wiki-ready knowledge.
6. In OpenCode API mode, install OpenCode locally, then refresh and choose the model and Agent in settings.

Safety Boundary

• Existing Raw files are not rewritten.
• Files are not deleted.
• Archiving is not automatic.
• Knowledge tasks write to the configured wiki, outputs, journal, index, and tracker paths.

Download Check

• Asset: obsidian-codex-0.4.0.zip
• SHA256: 154bbce9671070e75d25356d0f290960ce483289012ba034cf39df085bfaedc1

View on GitHub ↗

v0.3.0 - 写作上下文 Harness

这次把编辑区 改写 / 扩写 / 续写 从一次性文本请求升级成可见、可控的写作上下文 Harness。

新功能

• 新增 快速 / 质量 / 严格 三档写作质量模式。
• 新增侧栏 写作 状态入口和写作上下文面板。
• 新增结构化文章理解：主题、受众、写作目的、文章结构、关键事实、风格特征、禁止编造、局部写作建议。
• 新增文章理解软复用：小幅连续改写、扩写、续写会复用已有理解，不再每次重新理解全文。
• 新增严格模式审校：候选生成后再检查事实、风格、衔接、重复和 Markdown。
• 保留灰色候选闭环：Enter 确认，Esc 取消。
• 文章理解不进入普通聊天记录。

使用方法

1. 在插件设置里开启写作操作。
2. 选择默认写作质量：快速、质量 或 严格。
3. 在编辑区选中文字，运行 改写、扩写 或 续写。
4. 点击侧栏顶部 写作 状态，查看或刷新文章理解。
5. 按 Enter 接受灰色候选，或按 Esc 取消。

下载校验

• 安装包：obsidian-codex-0.3.0.zip
• SHA256：9ab1b80a4ebe96ddf6d82e12b0a0a06d51d0bbf28577f75e64df08af58e3538a

View on GitHub ↗

v0.2.0

Bug 修复

• 修复 Codex 账号重新登录后，插件可能因为找不到 Codex Desktop 内置 CLI 而报 spawn codex ENOENT 的问题。
• 设置页新增“刷新登录状态”，用于手动重连 Codex 并读取最新登录态。

实验功能

• 新增编辑区选中文字后的 改写、扩写、续写。
• 返回后在原地显示灰色候选，按 Enter 接受，按 Esc 取消。
• 该功能仍处于实验阶段，默认关闭，不成熟，不建议日常稳定使用。

测试方法

1. 在插件设置里开启写作操作。
2. 在编辑区选中文字，右键选择 改写、扩写 或 续写。
3. 先在非关键笔记里测试。

下载校验

• 安装包：obsidian-codex-0.2.0.zip
• SHA256：6642be3fd762973fee38e61e6b04e07c9ccc63d8a50daf2d9ffa7895b9a5c6d6

View on GitHub ↗

v0.1.2

新功能

公开发布内容保护，GitHub 仓库只保留安装和使用必要内容。

新功能使用说明

1. 下载最新 Release 安装包。
2. 安装 obsidian-codex 插件文件夹。
3. 直接使用插件。

View on GitHub ↗

v0.1.1

新功能

在 Codex 输入框里直接粘贴微信截图或系统截图。

使用方法

1. 截图。
2. 点击 Codex 输入框。
3. 按 Command+V，然后发送。

View on GitHub ↗

首个可分享版本。下载 obsidian-codex-0.1.0.zip，解压后把 obsidian-codex 文件夹放进 Obsidian vault 的 .obsidian/plugins/ 目录，然后在 Obsidian 第三方插件中启用。

View on GitHub ↗