Repository navigation
Releases
Releases ยท LibreChat-AI/LibreChat
v0.8.9-rc1
https://www.librechat.ai/changelog/v0.8.9-rc1
๐๏ธ Highlights
- Chat Projects with shared instructions and reference files
- Passkey sign-in, email changes, and stronger two-factor account protection
- Redesigned chat controls, reply notifications, and right-to-left layouts
- Deployment themes with configurable colors and appearance
- MCP Apps and more deliberate tool-approval choices
- Attached-machine enrollment, per-chat workspace choices, attachment copies, and pull request status
- Langfuse prompts in the catalog and slash command
- Recoverable subagent transcripts and clearer background-task progress
- Node.js 24.22.1 security updates and bounded history/tool previews
Upgrade guidance: https://www.librechat.ai/docs/configuration/upgrade
Click to expand Release Notes
โจ Features
- โฌ๏ธ feat: Migrate to Zod v4 by @danny-avila in #17442
- ๐ feat: Filter Agent Builder Selector to Authored Agents by @berry-13 in #17439
- ๐ฟ feat: Branch Tool Fold Rails Into Each Row by @berry-13 in #17407
- ๐ด feat: Use Langfuse Prompts from Cards and the Slash Command by @AtefBellaaj in #17104
- ๐ฌ feat: Sync Langfuse Prompts into the Catalog by @AtefBellaaj in #17095
- ๐๏ธ feat: Make External Prompt Groups Read-Only by @AtefBellaaj in #17008
- โฒ feat: Resolve Native and Langfuse Prompts by Source by @AtefBellaaj in #16909
- ๐บ feat: Track Prompt Group Source Identity by @AtefBellaaj in #16891
- ๐ง feat: Preview Langfuse Prompts in Settings by @AtefBellaaj in #16854
- ๐ช feat: Add Shared Source Author and Distributed Lock Helpers by @AtefBellaaj in #17086
- ๐งฌ feat: Re-Derive Azure Groups, Vertex AI, Custom Parameters and Spec Defaults for Config Profiles by @danny-avila in #17399
- ๐ท๏ธ feat: Hide and Rename Custom Endpoints by Name in Config Profiles by @danny-avila in #17396
- ๐งพ feat: Admin Config Write Report, Structured Issues and Deployment-Only Paths by @danny-avila in #17395
- ๐ญ feat: Admin Config Read APIs (Resolved and Effective Config, Scope Catalog, Env Status) by @danny-avila in #17383
- ๐ feat: Copy Chat Attachments into Attached Workspaces by @danny-avila in #17376
- ๐ feat: Admin Config Secrets v2 (MCP and Azure Credentials, Re-entry, Env Allowlist, Rebinding) by @danny-avila in #17353
- ๐งโ
โ๏ธ feat: Enroll Machines Through Explicit Browser Approval by @lia-by-librechat[bot] in #17344 - ๐ feat: Let Config Profiles Hide librechat.yaml MCP Servers and Remove Their Fields by @danny-avila in #17327
- โ๏ธ feat: Atomic Admin Config Changes Endpoint (set, unset, tombstone, entry upsert/remove/rename) by @danny-avila in #17328
- ๐ feat: Link Custom Artifact Mode to Its Instructions Guide by @danny-avila in #17021
- ๐ feat: Keep Dotted Record Keys in Admin Config Writes and Publish the Config Contract by @danny-avila in #17322
- feat: add masked session replay lifecycle by @graphaelli in #17194
- ๐ feat: Keyboard Submit and Escape for Tool Approval Cards by @berry-13 in #17233
- ๐ feat: Defer MCP OAuth Sign-In Until a Deferred Tool Is Used by @berry-13 in #17225
- ๐จ feat: Paint Inline Deployment Themes From the First Frame by @berry-13 in #17203
- ๐ข feat: Add Microsoft, Z.ai, IBM and Amazon Nova Model Marks by @berry-13 in #17250
- ๐ชช feat: Show Each Model's Maker on Its Model List Row by @berry-13 in #17246
- ๐งฌ feat: Show the Model Maker's Icon When Another Host Serves the Model by @berry-13 in #17238
- โจ๏ธ feat: Keyboard Submit, Focus and Escape for Ask User Cards by @berry-13 in #17231
- ๐ feat: Add Gemini 4 Argon Support by @berry-13 in #17200
- ๐ feat: Add Claude Haiku 5.5 Support by @berry-13 in #17201
- ๐๏ธ feat: Open Subagents Directly from Background Tasks by @lia-by-librechat[bot] in #17020
- ๐๏ธ feat: Make Context Usage Responsive and Condensed by @lia-by-librechat[bot] in #17018
- ๐ feat: Integrate Durable Skill Lifecycle Across Writers and Routes by @dustinhealy in #17043
- ๐งน feat: Add Bounded Whole-Skill Cleanup Engine by @dustinhealy in #17042
- ๐งพ feat: Add Skill Upload Reservation Helpers by @dustinhealy in #17041
- ๐งฑ feat: Add Skill Lifecycle Persistence Primitives by @dustinhealy in #17040
- ๐๏ธ feat: Expose Skill Deletion in Management API by @dustinhealy in #17005
- ๐งฉ feat: Add Skill Creation and File Deletion to Management API by @dustinhealy in #16878
- ๐๏ธ feat: Audit MCP, Code Worker, Langfuse, Skill Credential, Passkey and Audit Log Access by @berry-13 in #17054
- ๐ feat: Audit Resource Sharing Changes and Fail Closed on Reversible Audit Writes by @berry-13 in #17053
- ๐ feat: Audit Admin Role, Group, Config and IdP Role Mutations by @berry-13 in #17052
- ๐ feat: Cap the Select List at the Popover's Available Height by @berry-13 in #16938
- ๐จ feat: Paint the Operator Theme on a First-Ever Visit by @berry-13 in #16962
- ๐ feat: Optional Greeting Heading and Footer-Free Welcome Screen by @berry-13 in #17055
- ๐ feat: Hand the Colour Mode Over on an Entry Link by @berry-13 in #17056
- ๐ feat: Add a Connected Account Setting for Pull Requests by @berry-13 in #16990
- ๐ณ๏ธ feat: Approval-Responded, Output-Denied and Ask-User Parts in the UI Message View by @berry-13 in #16961
- ๐ feat: Let Users Connect Their Own Account for Pull Request Lookups by @berry-13 in #16989
- โป๏ธ feat: Share Pull Request Lookups Across Replicas Through Redis by @berry-13 in #16988
- ๐ชฃ feat: Look Up Bitbucket Cloud Pull Requests by @berry-13 in #16987
- โจ๏ธ feat: Change Default Chat Preferences by [@berry-13](https://github.com/...
v0.8.8
https://www.librechat.ai/changelog/v0.8.8
๐๏ธ Highlights
- Meet Lia: Building LibreChat with LibreChat.
- Agent control: Interrupt, steer, queue messages, answer questions, and approve tool calls.
- Unified Agent Builder: Same-run Skill authoring, opt-in Programmatic Tool Calling, and background tools.
- Background tasks: Clearer task cards, header controls, and automatic delivery of supported results.
- Scheduled Chats: Cron, time zones, and Project destinations. In beta: experimental and disabled by default.
- Code workspaces: Stateful sessions, attached machines, recovery, and guarded permissions. Highly experimental; compatible services required.
- Beta Agents API: Improved compatibility with Chat Completions, Open Responses, Agent/Skill management, and introduced Agent Events
- Experimental Agent Plugins: Bundle Skills and MCP servers, with opt-in command hooks.
- Execution visibility: Live activity, tool timing, Context Usage, optional costs, and a configured Langfuse Trace Viewer.
- Chat improvements: Draggable pins, chat filtering, manual compaction, mobile polish, and automatic attachment routing.
- Model updates: GPT-6 and GPT-6.1 Sol, Claude Opus/Sonnet 5.5, and Gemini 3.8 Flash. GPT-6.1 Sol uses family-fallback limits and pricing.
- Performance and security: Faster rendering and streaming and long-thread sends, stronger credential boundaries, tenant-scoped endpoints, and 2FA management limits.
Click to expand Release Notes
For detailed changes in the release candidates, see:
Operator configuration changes: Config v1.3.15, v1.3.16, and v1.3.17.
Changes Since v0.8.8-rc4
โจ Features
- ๐๏ธ feat: Recover Coding Chats With Missing Workspaces by @lia-by-librechat in #16273
- ๐ซ๏ธ feat: Add Bounded PII Text Transformation (library groundwork only; message ingress still blocks matches) by @lia-by-librechat in #16295
- ๐ feat: View and Stop Background Tasks From the Chat Header by @danny-avila in #16348
- ๐งฒ feat: Attach or Leave a Workspace on a Saved Chat by @lia-by-librechat in #16124
- ๐ช feat: Title an Open Activity Fold and Reach Failed Calls in One Click by @danny-avila in #16390
- ๐ญ feat: Bring the Streaming Thought Peek Back Under a Collapsed Live Card by @danny-avila in #16394
- ๐ด feat: Render Background Tasks as Native Tool Cards by @lia-by-librechat in #16407
- ๐ผ feat: Add Claude Sonnet 5.5 Support by @danny-avila in #16460
- ๐ feat: Distinguish Tool Preparation From Tool Call Time by @lia-by-librechat in #16455
- ๐ค๏ธ feat: Route Linked Worktree Requests into Per-Worktree Lanes by @danny-avila in #16477
- ๐ฏ feat: Tolerant Workspace Edits with Host-Rendered Conflict Diagnostics by @danny-avila in #16520
- ๐งฌ feat: Resolve GPT Point Releases to Their Family and Add GPT-6.1 Sol by @danny-avila in #16526
- ๐๏ธ feat: Scope YAML Custom Endpoints by Tenant by @danny-avila in #16537
๐ Fixes
- ๐ซ fix: Skip Blank RAG Queries Before Embedding by @lia-by-librechat in #16231
- ๐ช fix: Restore Measured Width to Virtualized Model Lists by @lia-by-librechat in #16252
- ๐ fix: Keep Queue Controls with the Starting Turn by @lia-by-librechat in #16268
- ๐ fix: Sync Workspace Choices When Chats Start by @lia-by-librechat in #16266
- ๐ธ fix: Scope Queued-Turn Deletion to Selected Conversations by @lia-by-librechat in #16270
- ๐ fix: Constrain the Agent Selector to 480px and 10 Own Agents by @berry-13 in #16256
- ๐ค fix: Accept GitHub App bot commit identities by @danny-avila in #16253
- ๐ fix: Restore Bedrock GPT Context for Long Prompts by @lia-by-librechat in #16284
- ๐ฆบ fix: Skip Redundant MCP OAuth Probes and Teardown Cancellations by @lia-by-librechat in #16298
- ๐ช fix: Keep Workspace Discovery Reactive by @lia-by-librechat in #16296
- ๐ชด fix: Surface Persistent Memory Before First Save by @lia-by-librechat in #16301
- ๐ฅ fix: Keep Tool Glyphs Behind Approval Reviews by @lia-by-librechat in #16307
- ๐ฆ fix: Quiet Idle Subagent Discovery While Preserving Child Wakeups by @lia-by-librechat in #16308
- ๐ช fix: Count Large Prompts in Bounded Tokenizer Chunks by @lia-by-librechat in #16305
- ๐ช fix: Preserve Coding Approval Mode on Background Returns by @lia-by-librechat in #16322
- ๐ข fix: Back Off Waiting Completion Wake-ups and Keep Their Trace Out of Requests by @danny-avila in #16349
- โก fix: Deliver Waiting Completion Wake-ups the Moment They Are Ready by @danny-avila in #16339
- ๐๏ธ fix: Discover Managed Agents in Marketplace Search by @lia-by-librechat in #16324
- ๐ฃ fix: Surface MCP Credential Failures During Tool Loading by @lia-by-librechat in #16323
- ๐งท fix: Wait for Long Dispatch Turns Before Giving Up Background Code Results by @danny-avila in #16345
- ๐ฎ fix: Count Undelivered Background Results as Outstanding by @danny-avila in #16343
- ๐ฅ fix: Pause Rejected Steer Recoveries by @lia-by-librechat in #16306
- ๐ฑ fix: Carry Coding Approval Mode With Queued Turns by @lia-by-librechat in #16326
- ๐งท fix: Anchor Background Results Onto Tool Calls a Resumed Turn Saved Without Step IDs by @danny-avila in #16353
- ๐ณ fix: Fail the Docker Build When the Frontend Build Fails by @danny-avila in #16388
- ๐ง fix: Keep Manual Compactions on the Active Branch by @lia-by-librechat in #16386
- ๐ฌ fix: Show Background Results Still Waiting for the Agent by @danny-avila in #16363
- ๐ fix: Project Only Accepted Agents API Tool Calls by @lia-by-librechat in #16384
- ๐งจ fix: Reject Invalid AES Credentials Before Startup by @lia-by-librechat in [#16402](#1640...
v0.8.8-rc4
https://www.librechat.ai/changelog/v0.8.8-rc4
๐๏ธ Highlights
- GPT-6 Sol, GPT-6 Luna, and Claude Opus 5.5
- Caller-executed tools for the Open Responses API
- Same-run Skill authoring and execution
- Agentic Coding:- conversation-isolated attached workspaces + repository instructions (
AGENTS.md)
- conversation-isolated attached workspaces + repository instructions (
- Expanded Trace Viewer with steps, roles, previews, and cost powered by Langfuse
- System-event turns and single-row live activity
- Faster streaming Markdown, model search, and Agent messages
- Stronger MCP authentication and per-request headers
Click to expand Release Notes
โจ Features
- โจ feat: Client-Executed Function Tools for the Responses API by @alexkorep in #16078
- ๐๏ธ feat: Add Programming App to OpenRouter Categories by @lia-by-librechat in #16228
- ๐ feat: Add GPT-6 Sol and Luna Support by @lia-by-librechat in #16221
- ๐ feat: Support Cognito Access Tokens for Agent Management by @dustinhealy in #16224
- ๐น feat: Add Claude Opus 5.5 Support by @danny-avila in #16215
- โฒ๏ธ feat: Report a Background Task's Start, Settlement and Elapsed Time by @lia-by-librechat in #16193
- ๐งโ๐ feat: Add Grok 4.7 Support Through xAI by @lia-by-librechat in #16173
- ๐ซง feat: Hold a Live Run's Activity at One Row by @danny-avila in #16118
- ๐ feat: Render System Events as Distinct Turns by @danny-avila in #16119
- ๐ต feat: Show Cost per Record, Step and Response in the Trace Ledger by @danny-avila in #16120
- ๐ feat: Authenticate RUM Proxy Exports by @lia-by-librechat in #16114
- ๐ฃ feat: Invoke Newly Authored Skills in the Same Run by @lia-by-librechat in #16036
- ๐ณ feat: Isolate Attached Workspaces by Conversation by @danny-avila in #16107
- ๐ฌ feat: Read a Model Call as a Conversation in the Trace Viewer by @danny-avila in #16103
- ๐ฃ๏ธ feat: Trace Viewer Roles, Agent Identity and Tool Rounds by @danny-avila in #16098
- โฝ feat: Negotiate Background Command Timeouts by @danny-avila in #16100
- ๐ช feat: Trace Viewer Steps, Sequence Scale and Previews by @danny-avila in #16009
- ๐ช feat: Send Per-Request MCP Headers Without Hiding the Tool Catalog by @lia-by-librechat in #15988
- ๐ feat: OpenAPI spec and docs for the public Agents API by @AtefBellaaj in #15928
- ๐ Load Repository Instructions for Attached Workspaces by @danny-avila in #16008
- ๐ชช feat: Pass Schedule Identity to MCP Token Providers by @danny-avila in #15962
๐ Fixes
- ๐ฉผ fix: Keep OpenRouter Category Override Test Type-Safe by @lia-by-librechat in #16229
- ๐ fix: Enable File Search When Attaching From the Files Panel by @jtodaroii in #16223
- ๐ฝ fix: Give Each Streamed Tool Call a Stable Outward Index by @lia-by-librechat in #16209
- ๐ฐ fix: Show Live Tool Combo Multipliers by @lia-by-librechat in #16202
- โฒ fix: Prevent Agent Model Stream Idle Timeouts by @lia-by-librechat in #16201
- ๐ฅซ fix: Normalize Legacy MCP Header Maps on Cached Reads by @lia-by-librechat in #16200
- ๐ซฅ fix: Preserve Absent MCP Header Maps Across Database Saves by @lia-by-librechat in #16194
- ๐ด fix: Split Combined Bit Updates in ACL Permission Writes by @lia-by-librechat in #16172
- ๐ฅ fix: Preserve Insights With Guarded ACL Role Updates by @lia-by-librechat in #16170
- ๐ฆ fix: Require Named Helm Credential Secrets by @lia-by-librechat in #16175
- ๐ fix: Fill Reasoning Previews Before Throttling by @lia-by-librechat in #16178
- โน๏ธ fix: Cancel Stale Throttled Scroll Callbacks by @minwookshin in #16167
- ๐งป fix: Unblock Conversation Deletion After Queued Turns by @lia-by-librechat in #16174
- ๐ค fix: Validate MIME Patterns With the Active Regex Engine by @lia-by-librechat in #16177
- ๐ฏ fix: Explain Tenant Index Migration on Startup Conflicts by @lia-by-librechat in #16176
- ๐ fix: Preserve Chat Drafts Across Navigation and Clearing by @danny-avila in #16161
- ๐ fix: Reuse Redis Caches Without Changing TTLs by @lia-by-librechat in #16154
- ๐ช fix: Preserve Tool Pane Choices Through Response Finalization by @lia-by-librechat in #16152
- โ๏ธ fix: Persist a Generated Title Before the Turn Ends by @lia-by-librechat in #16137
- ๐ fix: Stop Painting the Closed Mobile Drawer by @lia-by-librechat in #16127
- ๐ fix: Preview SVG Artifacts From Their Edited Source by @lia-by-librechat in #16123
- ๐ผ๏ธ fix: Resolve Pinned Model Icons from Endpoint Config by @jtodaroii in #16089
- โ๏ธ fix: Render Code in the Bundled Monospace Font by @lia-by-librechat in #16146
- ๐ซง fix: Stop the Live Activity Row From Flashing Open on Code Calls and Thoughts by @danny-avila in #16145
- ๐ท fix: Isolate Jina Credential Provenance by @danny-avila in #16134
- ๐ชข fix: Bind Preserved MCP API Keys to Server Configuration by @danny-avila in #16136
- โ fix: Keep a Resumed Compaction Anchored to the Turn It Summarizes by @lia-by-librechat in #16037
- ๐ณ๏ธ fix: Repair Legacy Meili Cleanup Index by @danny-avila in #16133
- โ fix: Preserve MCP Authorization During Token Endpoint Outages by @lia-by-librechat in #16126
- ๐ฝ fix: Hand Same-Batch Code Calls the Files a Skill Just Loaded by @lia-by-librechat in #16115
- ๐ฅข fix: Exclude Prompt Inputs From Terminal Events by @lia-by-librechat in #16112
- ๐ฅ fix: Refresh MCP Credentials That Arrive Alread...
v0.8.8-rc3
https://www.librechat.ai/changelog/v0.8.8-rc3
๐๏ธ Highlights
- Experimental: Bring-Your-Own-Machine for scalable coding agents
- Agent/Skills CRUD API improvements
- Conversation Trace Viewer and Subagent file sharing
- GPT-6 Astra and expanded Context Usage
- File text fallback, unified artifact rows, and upload previews
- Unified pinning, chat filtering, and interface polish
- MCP, OpenID, Agent, and Code reliability improvements
Click to expand Release Notes
โจ Features
- ๐งท feat: Carry Ask User Answers Past Compaction by @danny-avila in #15944
- ๐ช feat: Opt-In Text Fallback for Files No Tool Can Read by @danny-avila in #15940
- ๐น๏ธ feat: Run Declared Project Actions on Attached Workers by @danny-avila in #15943
- ๐ชง feat: Unify Artifact Triggers as Tool Rows by @berry-13 in #15915
- ๐ feat: Retry OpenID Discovery After Startup Failures by @danny-avila in #15899
- ๐ญ feat: Conversation Trace Viewer by @danny-avila in #15869
- ๐ฆ feat: Share Run-Scoped Files With Subagents by @usnavy13 in #15848
- ๐ feat: Filter and Sort the Sidebar Chats List by @berry-13 in #15842
- ๐ feat: Record Safe Upstream Model Failures by @dustinhealy in #15800
- ๐ feat: Render Safe Upstream Model Errors by @dustinhealy in #15837
- ๐๏ธ feat: Seal Code Environment Choices Per Conversation by @danny-avila in #15868
- ๐ก feat: Remember Agent Workspace Defaults by @danny-avila in #15843
- ๐งฎ feat: Surface Retained Tool-Message Context Usage by @berry-13 in #15796
- ๐ฉป feat: improve agent filter controls for insights by @upman in #15807
- ๐ feat: Unified Pinned Section With Sidebar Drag Interactions by @berry-13 in #15057
- ๐ฅผ feat: Expose BYOM Cancellation Diagnostics by @danny-avila in #15810
- ๐ช feat: Cancel Background Tool Invocations by @danny-avila in #15773
- ๐ช feat: Expose Bounded Attached Bash Timeouts by @danny-avila in #15776
- ๐๏ธ feat: Select Attached Code Workspaces by @danny-avila in #15768
- ๐ feat: Route Attachments and Provision Tools Lazily by @danny-avila in #15763
- ๐๏ธ feat: Establish Agent Management API Contract by @danny-avila in #15545
- ๐ feat: Add M2M Authentication for Agent Management by @dustinhealy in #15555
- ๐ชต feat: Add OpenTelemetry Logs Pipeline via Winston Bridge by @danny-avila in #15565
- ๐ feat: Add Agent Management Read Endpoints by @dustinhealy in #15561
- โจ feat: add morphing state icons across the client by @berry-13 in #14697
- โ feat: Add Agent Management Creation Endpoint by @dustinhealy in #15564
- โ๏ธ feat: Add Agent Management Update Endpoint by @dustinhealy in #15566
- ๐ช feat: Add Tenant-Scoped Agent Deletion by @dustinhealy in #15573
- ๐ feat: Read Attached Workspace Files by @danny-avila in #15524
- ๐ฏ๏ธ feat: Search Text in Attached Workspaces by @danny-avila in #15525
- ๐ feat: Add GPT-6 Astra Support by @danny-avila in #15567
- ๐จ feat: Theme-Adaptive SVG Support for Custom MCP and Group Icons by @berry-13 in #13817
- ๐ณ feat: List Attached Workspace Trees by @danny-avila in #15527
- ๐๏ธ feat: Provision Native Code Workers by @danny-avila in #15587
- ๐ป feat: Route Bash Through Attached Workspaces by @danny-avila in #15546
- ๐ feat: Author Files in Attached Workspaces by @danny-avila in #15548
- ๐ฐ๏ธ feat: Show BYOM Worker Readiness by @danny-avila in #15618
- ๐ฉป feat(insights): Scope Insights to Authorized Agents by @upman in #15549
- ๐ feat: Copy Messages as Rich Text by @berry-13 in #15087
- โฟ feat: High Contrast Light and Dark Appearance Modes by @berry-13 in #15178
- ๐ชช feat: Add Per-Agent Git Identity by @danny-avila in #15628
- ๐ญ feat: Tag Browser Diagnostics With Client Build IDs by @danny-avila in #15656
- ๐งถ feat: Flatten Active Message Threads Behind a Flag by @danny-avila in #15633
- ๐ feat: Add Agent File Management APIs by @danny-avila in #15666
- ๐ feat: Add AMD Lemonade Provider Recognition and Branding by @berry-13 in #15664
- ๐งฐ feat: Unify Tool Activity, Reasoning, Search, and Agent Workflows by @berry-13 in #14546
- ๐ feat: Add Configurable BYOM Enrollment Policy by @danny-avila in #15673
- ๐ feat: Add Skills Discovery and Authoring to the Management API by @danny-avila in #15642
- ๐ชฌ feat: Configure Langfuse Trace User Identity and Metadata by @danny-avila in #15667
- ๐ณ๏ธ feat: Define Conversation Code Approval Modes by @danny-avila in #15685
- ๐งถ feat: Thread Conversation Code Approval Through Requests by @danny-avila in #15698
- ๐ชง feat: Surface Conversation Code Approval Modes by @danny-avila in #15703
- ๐ช feat: Compact Context in a Summarize-Only Turn by @danny-avila in #15690
- ๐ง feat: Restore Interactive Tool Approval Review by @danny-avila in #15717
- ๐ซ feat: Enable Full Access for BYOM Coding by @danny-avila in #15718
- ๐งฐ feat: Move Code Settings Into Run Code by @danny-avila in #15722
๐ Fixes
- ๐งผ fix: Render Agent Description HTML Safely by @danny-avila in #15951
- ๐ซด fix: Preserve Queued Follow-Ups Through Handoff by [@...
v0.8.8-rc2
https://www.librechat.ai/changelog/v0.8.8-rc2
๐๏ธ Highlights
- Interrupt Agent runs before visible output, steer or queue follow-ups, and recover tool-limited turns
- Optional generated activity groups, live phase cards, and reasoning labels
- Unified Agent Builder with standalone Skill authoring, background tools, intent labels, and Programmatic Tool Calling
- Durable Agent Events, actor mailboxes, receipts, human pauses, and background completion wakeups
- Saved Agent teams, deeper Subagent history, stable event panels, live controls, and automatic parent continuation
- Experimental Agent Plugins and highly experimental Code environments with guarded file and command permissions
- Agent memory isolation and categorized Context Usage with optional cost
- Redesigned Projects, full-message chat search, searchable settings, shortcuts, pinning, and mobile navigation
- Experimental Scheduled Chats with cron, time zones, weekly cadence, and project destinations
- Stable shared links, personal copies, editable long pastes, and attachment-only turns
- Fullscreen artifacts, Mermaid export, PowerPoint templates, and original file downloads
- Keenable keyless web search plus expanded SearXNG and Tavily controls
- Default security headers, opt-in nonce CSP, authenticated local images, Code Interpreter JWTs, and stronger SAML/OIDC flows
- In-app Langfuse connections, tenant fanout, export telemetry, and authorized session links
- Auditable source-aware content filtering, tenant Insights, and encrypted admin configuration
- GPT-5.6, Claude Fable 5.1, Opus 5 and Sonnet 5, Gemini 3.8/3.7/3.6 Flash, and Gemini 3.5 Flash-Lite
- Adaptive streaming, automatic generation protocol v2, Redis failover recovery, and live MCP catalog refresh
Click to expand Release Notes
โจ Features
- ๐ feat: Track terms acceptance timestamp by @berry-13 in #10810
- ๐ feat: Continue Shared Conversations as Personal Copies by @berry-13 in #13714
- ๐ช feat: HITL Tool Approval Scaffolding (Slice A) by @danny-avila in #12938
- ๐ง feat: Add Memory as an Agent Capability with Inline Tools and Ephemeral Badge by @danny-avila in #13869
- ๐ feat: Agent Contact Visibility with Owner Fallback by @peeeteeer in #13663
- ๐ช feat: Faithful Over-Window Context Estimate via Prune Mirror and Overhead Reserve by @danny-avila in #13959
- ๐ชญ feat: Add opt-in Langfuse fanout gateway + collector by @upman in #13872
- ๐ช feat: Human-in-the-Loop Runtime - Tool Approval + Ask-User-Question (Slice B) by @danny-avila in #13942
- ๐ช feat: Dock-Style Fisheye Nav Rail With Instant Hover Preview by @danny-avila in #14021
- ๐ค feat: Add Claude Sonnet 5 Support by @danny-avila in #14042
- ๐ฅท feat: Add
showInMenuOption to Model Specs by @alexkorep in #14034 - ๐ซท feat: Exclude File Authoring Tools From Eager Execution by @danny-avila in #14051
- ๐ช feat: Configurable Tool-Approval Policy via Programmatic Hooks by @danny-avila in #14025
- ๐งน feat: Eager HITL Checkpoint Cleanup (Expiry + Deletion) & Full-Wiring E2E by @danny-avila in #14123
- ๐งฐ feat: Redesign Agent Builder with Unified Tools Marketplace, Skills & Orchestration by @berry-13 in #13952
- ๐ก feat: add rum browser page-load diagnostics by @upman in #14106
- ๐๏ธ feat: Explicit "Use All Skills" Toggle in the Skills Section by @berry-13 in #14125
- ๐ feat: ask_user_question - agent-initiated questions with durable pause/resume by @danny-avila in #14139
- ๐๏ธ feat: Per-Agent Memory Partitions by @danny-avila in #14084
- ๐ค feat: Add GPT-5.6 (Sol/Terra/Luna) OpenAI Models by @danny-avila in #14206
- ๐ชญ feat: support per-run central export suppression by @upman in #14207
- โจ feat: stateful_code_sessions capability for warm Code API sandbox sessions (experimental) by @danny-avila in #14150
- ๐ง feat: Add GPT-5.6 reasoning.mode + reasoning.context (Responses API) by @danny-avila in #14233
- ๐งต feat: Background Tool Calls for Agents & Model Specs by @danny-avila in #14197
- ๐๏ธ feat: Encrypted Langfuse Fanout Config by @dustinhealy in #14107
- ๐งญ feat: Mid-Run Steering and Queued Messages for Agent Runs by @danny-avila in #14220
- โจ๏ธ feat: Prewarm Stateful Code Sandboxes with Cold-Boot UX Feedback by @danny-avila in #14239
- ๐ feat: Add Client Name Header to Tavily Search Requests by @manisrinivasan2k1 in #14082
- ๐ฌ feat: Explain Steering & Queuing at Every User-Facing Surface by @danny-avila in #14260
- ๐ผ๏ธ feat: Return Sandbox Images From
read_fileas Viewable Artifacts by @danny-avila in #14277 - feat: add low-noise Redis observability by @upman in #14309
- โ feat: Add Reclaim-Gated Controls to In-Flight Steers by @danny-avila in #14321
- ๐ซง feat: Float In-Flight Steers Over the Thread with Collapsible Content by @danny-avila in #14365
- โก feat: Add Gemini 3.6 Flash and Gemini 3.5 Flash-Lite Support by @danny-avila in #14369
- ๐งต feat: Native Background Execution for Code Interpreter Tools by @danny-avila in #14386
- โณ๏ธ feat: Claude Opus 5 Support by @danny-avila in #14422
- โฑ๏ธ feat: Configurable HTTP Server Timeouts by @danny-avila in #14481
- ๐ถ feat: Generalize Admin Config Secret Redaction by @dustinhealy in #14509
- ๐ท๏ธ feat: Activity Groups With Fast-Model Headers by @danny-avila in #14391
- ๐๏ธ feat: Custom Endpoint API Key Encryption via Array Secret Registry by @danny-avila in #14510
- ๐ฏ feat: Tool Intent Label Capability (tool_intents) by @danny-avila in #14499
- ๐ชข feat: Langfuse Fanout Connection Setting by @danny-avila in #14108
- ๐ feat: Preemptive Steer - Backend Interrupt & Steer by @danny-avila in #14518
- ๐ฏ feat: Per-Tool Intent Labels for Model Specs by @danny-avila in #14526
- โก feat: Interrupt & Steer, Client by @danny-avila in #14528
- ๐งต feat: Background-Native Code Execution Tools by @danny-avila in #14532
- ๐ฏ feat: Render Tool Intent as the Live Tool-Call Label by @danny-avila in #14536
- ๐ฏ feat: Per-Tool Intent Label Toggles in the Agent Builder by @danny-avila in #14550
- โก feat: Reliable Interrupt & Steer Escalation and Recovery by @danny-avila in #14558
- ๐ฌ feat: Interim Progress Card for Streaming Q&A Calls by @danny-avila in #14576
- ๐ฆ feat: Configurable Circuit Breakers for Runaway Streamed Tool Args by @danny-avila in #14613
- โก feat: Coalesce Redis Streaming Delta Publications into Windowed Batches by @danny-avila in #14614
- ๐ feat: Add allowedAddresses Exemption to Speech (STT/TTS) and OCR Config Schemas by @dustinhealy in #14559
- ๐ feat: Wire Adaptive Stream Smoothing Across Google, Bedrock, and Zero-Disable Semantics by @danny-avila in #14660
- ๐ feat: Admin Panel Link in Settings for Admins by @dustinhealy in #14662
- ๐ feat: Day-Aware Landing Greeting Schedule by @berry-13 in #14633
- ๐ feat: Refresh-Capable Google Admin OAuth Sessions by @dustinhealy in #13832
- ๐ก๏ธ feat: Add Batched MCP Authority Proofs by @danny-avila in #14688
- ๐ช feat: Add Claude Hook Compatibility Layer by @danny-avila in #14412
- โถ feat: add fullscreen artifact ...
v0.8.8-rc1
https://www.librechat.ai/changelog/v0.8.8-rc1
๐๏ธ Highlights
- Interrupt or steer Agent runs in progress, or queue the next message
- Agents can ask up to four related questions in one form, pause for input or approval, then resume
- Redesigned Agent Builder with a unified Tools marketplace and background tools
- Experimental Agent Plugins with deployment Skills, MCP servers, and opt-in command hooks
- Activity groups, parent phase summaries, and live tool intent labels
- Experimental stateful Code Interpreter sessions with reusable conversation workspaces
- Agent-managed memory with optional per-agent isolation
- Stable shared-link updates, active-link badges, and personal copies
- Fullscreen artifact previews and Mermaid SVG/PNG export
- In-app Langfuse connections, tenant fanout, and admin session links
- Unified message layouts, multi-part response editing, and full-message copy
- Smooth text fade-in, adaptive streaming, and live MCP tool-catalog refresh
- Stronger SSRF guards for speech, OCR, and web tools
- Provider updates: GPT-5.6, Claude Opus 5 and Sonnet 5, Gemini 3.7 and 3.6 Flash, and Gemini 3.5 Flash-Lite
Click to expand Release Notes
โจ Features
- ๐ feat: Track terms acceptance timestamp by @berry-13 in #10810
- ๐ feat: Continue Shared Conversations as Personal Copies by @berry-13 in #13714
- ๐ช feat: HITL Tool Approval Scaffolding (Slice A) by @danny-avila in #12938
- ๐ง feat: Add Memory as an Agent Capability with Inline Tools and Ephemeral Badge by @danny-avila in #13869
- ๐ feat: Agent Contact Visibility with Owner Fallback by @peeeteeer in #13663
- ๐ช feat: Faithful Over-Window Context Estimate via Prune Mirror and Overhead Reserve by @danny-avila in #13959
- ๐ชญ feat: Add opt-in Langfuse fanout gateway + collector by @upman in #13872
- ๐ช feat: Human-in-the-Loop Runtime - Tool Approval + Ask-User-Question (Slice B) by @danny-avila in #13942
- ๐ช feat: Dock-Style Fisheye Nav Rail With Instant Hover Preview by @danny-avila in #14021
- ๐ค feat: Add Claude Sonnet 5 Support by @danny-avila in #14042
- ๐ฅท feat: Add
showInMenuOption to Model Specs by @alexkorep in #14034 - ๐ซท feat: Exclude File Authoring Tools From Eager Execution by @danny-avila in #14051
- ๐ช feat: Configurable Tool-Approval Policy via Programmatic Hooks by @danny-avila in #14025
- ๐งน feat: Eager HITL Checkpoint Cleanup (Expiry + Deletion) & Full-Wiring E2E by @danny-avila in #14123
- ๐งฐ feat: Redesign Agent Builder with Unified Tools Marketplace, Skills & Orchestration by @berry-13 in #13952
- ๐ก feat: add rum browser page-load diagnostics by @upman in #14106
- ๐๏ธ feat: Explicit "Use All Skills" Toggle in the Skills Section by @berry-13 in #14125
- ๐ feat: ask_user_question - agent-initiated questions with durable pause/resume by @danny-avila in #14139
- ๐๏ธ feat: Per-Agent Memory Partitions by @danny-avila in #14084
- ๐ค feat: Add GPT-5.6 (Sol/Terra/Luna) OpenAI Models by @danny-avila in #14206
- ๐ชญ feat: support per-run central export suppression by @upman in #14207
- โจ feat: stateful_code_sessions capability for warm Code API sandbox sessions (experimental) by @danny-avila in #14150
- ๐ง feat: Add GPT-5.6 reasoning.mode + reasoning.context (Responses API) by @danny-avila in #14233
- ๐งต feat: Background Tool Calls for Agents & Model Specs by @danny-avila in #14197
- ๐๏ธ feat: Encrypted Langfuse Fanout Config by @dustinhealy in #14107
- ๐งญ feat: Mid-Run Steering and Queued Messages for Agent Runs by @danny-avila in #14220
- โจ๏ธ feat: Prewarm Stateful Code Sandboxes with Cold-Boot UX Feedback by @danny-avila in #14239
- ๐ feat: Add Client Name Header to Tavily Search Requests by @manisrinivasan2k1 in #14082
- ๐ฌ feat: Explain Steering & Queuing at Every User-Facing Surface by @danny-avila in #14260
- ๐ผ๏ธ feat: Return Sandbox Images From
read_fileas Viewable Artifacts by @danny-avila in #14277 - feat: add low-noise Redis observability by @upman in #14309
- โ feat: Add Reclaim-Gated Controls to In-Flight Steers by @danny-avila in #14321
- ๐ซง feat: Float In-Flight Steers Over the Thread with Collapsible Content by @danny-avila in #14365
- โก feat: Add Gemini 3.6 Flash and Gemini 3.5 Flash-Lite Support by @danny-avila in #14369
- ๐งต feat: Native Background Execution for Code Interpreter Tools by @danny-avila in #14386
- โณ๏ธ feat: Claude Opus 5 Support by @danny-avila in #14422
- โฑ๏ธ feat: Configurable HTTP Server Timeouts by @danny-avila in #14481
- ๐ถ feat: Generalize Admin Config Secret Redaction by @dustinhealy in #14509
- ๐ท๏ธ feat: Activity Groups With Fast-Model Headers by @danny-avila in #14391
- ๐๏ธ feat: Custom Endpoint API Key Encryption via Array Secret Registry by @danny-avila in #14510
- ๐ฏ feat: Tool Intent Label Capability (tool_intents) by @danny-avila in #14499
- ๐ชข feat: Langfuse Fanout Connection Setting by @danny-avila in #14108
- ๐ feat: Preemptive Steer - Backend Interrupt & Steer by @danny-avila in #14518
- ๐ฏ feat: Per-Tool Intent Labels for Model Specs by @danny-avila in #14526
- โก feat: Interrupt & Steer, Client by @danny-avila in #14528
- ๐งต feat: Background-Native Code Execution Tools by @danny-avila in #14532
- ๐ฏ feat: Render Tool Intent as the Live Tool-Call Label by @danny-avila in #14536
- ๐ฏ feat: Per-Tool Intent Label Toggles in the Agent Builder by @danny-avila in #14550
- โก feat: Reliable Interrupt & Steer Escalation and Recovery by @danny-avila in #14558
- ๐ฌ feat: Interim Progress Card for Streaming Q&A Calls by @danny-avila in #14576
- ๐ฆ feat: Configurable Circuit Breakers for Runaway Streamed Tool Args by [@danny-avila](https://github.com/danny...
v0.8.7
https://www.librechat.ai/changelog/v0.8.7
๐๏ธ Highlights
- Chat projects for organizing chats into folders
- Context Usage gauge showing current window usage, tokens used, and cost (optional)
- Agents with skills enabled can author new skills on the fly
- GitHub sync for skills, and local bundled skills
- Settings redesign, keyboard shortcuts, default pinned tools, and pinned conversations.
- Provider updates:
- Claude Fable 5, Anthropic custom endpoints, prompt cache TTL, Google URL Context with YouTube handling, and custom endpoint option for reasoning replay.
- Shared links, OpenID, and MCP now have stronger access controls and safer auth flows.
- Bundled Admin Panel to default compose files
- Released Code Interpreter for secure code execution and file generation
Click to expand Release Notes
For detailed changes in the release candidate, see:
Changes Since v0.8.7-rc1
โจ Features
- ๐ feat: Add Bottom Terminus Node to Message Minimap Navigation by @danny-avila in #13853
- ๐ช feat: SDK-Aligned Context-Usage Projection by @danny-avila in #13801
- ๐ feat: Add Pinned Conversations by @dlew in #13492
- ๐ feat: Tag Langfuse Traces With Tenant ID by @upman in #13808
- ๐ฐ๏ธ feat: Resolve Agent Prompt Time Variables in User's Timezone by @berry-13 in #13815
- ๐๏ธ feat: Redesign Settings with Registry-Driven Dialog, Search, and Mobile Drill-In by @berry-13 in #13722
- ๐ feat: Agent API Keys management UI in Settings โ Data controls by @berry-13 in #13819
- ๐ feat: Audit Log Backend for SystemGrant Assign and Revoke Events by @dustinhealy in #13087
- ๐ feat: Add
promptCacheTtlmodel parameter for 1h/5m cache duration by @danny-avila in #13835 - ๐ feat: Surface User-Provided API Keys in Settings, Scoped to Reachable Endpoints by @danny-avila in #13864
- ๐ช feat: Add Token Pricing for Devstral, Voxtral, Holo2, and Mistral Medium by @JumpLink in #13863
- ๐ feat: Seed Default Pinned Tools and MCP Dropdown via Interface Config by @danny-avila in #13865
- ๐ feat: Snapshot Files for Shared-Link Attachments by @danny-avila in #13740
- ๐๏ธ feat: Reference Selected Chat Text with Multi-Quote Popup by @danny-avila in #13868
- ๐ณ feat: Bundle Admin Panel in Docker Compose Stacks by @danny-avila in #13876
- โจ๏ธ feat: Keyboard Shortcuts by @berry-13 in #12425
- ๐ง feat: Configurable Reasoning Replay for Custom Endpoints by @danny-avila in #13921
- ๐บ feat: Google URL Context Param with Native YouTube Video Understanding by @danny-avila in #13924
๐ Fixes
- ๐ค fix: Keep Microphone Icon Visible On Initial Chat Render by @danny-avila in #13788
- ๐๏ธ fix: Support Windows ZIP MIME Uploads by @danny-avila in #13794
- ๐ฅฝ fix: Restrict MCP Server URL Disclosure to Admins, Owners, and Editors by @dustinhealy in #13784
- ๐ฉน fix: Bill Anthropic Prompt-Cache Tokens Once by @danny-avila in #13798
- ๐ช fix: Restore Raw Spec Fallback for Enforced Presets by @danny-avila in #13804
- ๐ fix: Honor Admin-Panel MCP Allowlist Overrides Without Restart by @danny-avila in #13814
- ๐ fix: Decrement Bookmark Counts When Deleting Conversations by @danny-avila in #13830
- ๐ fix: Surface Share Permissions Load Error as Alert Button With Tooltip by @danny-avila in #13833
- ๐งท fix: Preserve Document Priority on Section-Scoped Config Writes by @dustinhealy in #13772
- ๐ช fix: Match Prompt Cache TTL Control to Region Combobox Styling by @danny-avila in #13839
- ๐ fix: Resolve dayjs plugin ESM imports in data-provider by @danny-avila in #13851
- โ fix: Strip Remaining Unsupported JSON Schema Keywords for Gemini MCP Tools by @danny-avila in #13850
- ๐ fix: Apply OCR Context to Responses API Agents and Handoffs by @Airamhh in #13707
- ๐ท๏ธ fix: Scope File Search entity_id to Agent Knowledge-Base Files Only by @msserpa in #13693
- ๐ชข fix: Paginate MCP
tools/listto Load All Tools by @TomasPalsson in #13840 - ๐ช fix: Gate Artifacts Toggle on Agent Capability Flag by @serhiizghama in #13665
- ๐ฏ fix: Narrow Public Share 401 Bypass to the Share Endpoint Only by @danny-avila in #12905
- ๐ fix: Gate Shared Startup Config By Link Access by @danny-avila in #13897
- โ๏ธ fix: Cap Audit Chain Verification and Honor Client Cancellation by @danny-avila in #13903
- ๐ช fix: Validate Shared-File Cookie Auth Against the Live Refresh Session by @danny-avila in #13908
- ๐ฐ fix: Bound MCP
tools/listPagination with Aggregate Budgets by @danny-avila in #13909 - ๐ชฃ fix: Cap Context Projection Workload Before Tokenization by @danny-avila in #13910
- ๐ช fix: Strip Caller-Supplied Priority from Assign-Only Config Upserts by @danny-avila in #13911
- ๐ fix: Re-Check
execute_codeAuthorization on Event-Driven Tool Loads by @danny-avila in #13912 - ๐ค fix: Withhold Custom Endpoint Headers for User URLs by @danny-avila in #13917
- ๐งพ fix: Harden Historical File Authorization by @danny-avila in #13918
- ๐งญ fix: Harden User Provided Endpoint URL Protection by @danny-avila in #13919
- ๐ fix: Prevent Infinite Render Loop on Code-Execution File Preview by @danny-avila in #13922
- ๐ฑ๏ธ fix: Summon Quote Popup on Double-Click Word Selection by @danny-avila in #13923
- ๐งน fix: Order Enabled Guard After Config Expansion in
useGetAgentByIdQueryby @dlew in #13927 - ๐ต fix: Align Mention Empty Result Behavior With Skills Command by @Oliver777int in #13928
- ๐ชค fix: Guard Prompts Popover Against Empty Result Keyboard Navigation by @danny-avila in [#13...
v0.8.7-rc1
๐๏ธ Highlights
- Access, Auth, and MCP: Shared-link ACLs, OpenID role sync/token reuse, MCP OBO, request-scoped MCP flows, resumable cleanup, and OAuth/proxy hardening.
- Agents, Skills, and Projects: Chat projects, skill file authoring tools, deployment/GitHub Skills, scoped sync status, model-spec controls/starters, and marketplace access.
- Safety and Admin: Configurable message PII filtering and hardened model-spec Subagent controls.
- Observability and Operations: Auth-isolated RUM proxying with outcome metrics, auth fallback telemetry, opt-in Redis command tracing, Langfuse scores, Node 24 Docker builds, and stronger CI.
- Conversation UX: Immediate titles, hover timestamps, message-nav scrubbing, stable branches, sanitized shares, deploy recovery, and route reload fixes.
- Providers and Reasoning: Claude Fable 5, GPT-5.5/frontier defaults, native Anthropic custom endpoints, protected user URLs, provider headers, and Gemini/Imagen/Bedrock fixes.
- Streaming and Performance: Memoized Markdown streaming, per-tool
onResult, real-time context, opt-in cost display, scoped token config caches, subagent billing, lazy loading, andtsdownbuilds.
Click to expand Release Notes
What's Changed
โจ Features
- ๐ feat: Use SecretInput for Sensitive Fields by @berry-13 in #12955
- ๐ก feat: Add Authenticated Proxy Mode for Browser RUM Telemetry by @upman in #13464
- ๐ feat: Add On-Behalf-Of (OBO) token exchange support for MCP server connections by @jcbartle in #13429
- ๐ช feat: Add OpenID Role Sync by @danny-avila in #13415
- ๐๏ธ feat: Retain Agent Files During All-Data Retention by @danny-avila in #13477
- ๐ญ feat: Add Credential-Free Playwright Smoke Suite with a Local Mock LLM by @teresama in #13472
- โก feat: Immediate Conversation Title Generation by @danny-avila in #13395
- ๐ฐ๏ธ feat: Add Auth Fallback Observability by @danny-avila in #13488
- ๐ feat: Add Granular Access Control to Shared Links via ACL System by @AtefBellaaj in #13051
- ๐๏ธ feat: Add Private Chat Projects by @danny-avila in #13467
- ๐๏ธ feat: Add Focus Management and Drag-to-Scrub to MessageNav by @danny-avila in #13497
- ๐๏ธ feat: Add Agent File Authoring Tools by @danny-avila in #13435
- ๐ feat: Improve Skill Authoring Guidance by @danny-avila in #13517
- ๐งญ feat: Scope Model Spec Skills by @danny-avila in #13522
- ๐๏ธ feat: Add Deployment Skill Directory by @danny-avila in #13523
- ๐ช feat: Surface Agent Marketplace in Model Selector by @danny-avila in #13553
- ๐ฑ feat: Support Soft Default Model Spec by @danny-avila in #13554
- โณ feat: Make OpenID Token Reuse Window Configurable by @pboers1988 in #13546
- ๐ feat: Surface Message Feedback as Langfuse Scores by @graphaelli in #13544
- ๐งฉ feat: Enable Model Spec Subagents by @danny-avila in #13598
- ๐ก๏ธ feat: Configurable Message PII Filter by @dustinhealy in #13602
- ๐ feat: Add Claude Fable 5 Support by @danny-avila in #13628
- ๐ฐ๏ธ feat: Add GPT-5.5 + Frontier OpenAI Models, Drop Deprecated Defaults by @danny-avila in #13636
- ๐งฌ feat: Add GitHub Skill Sync by @danny-avila in #13293
- ๐ท๏ธ feat: Surface Model Spec Branding on Landing and Selector by @danny-avila in #13662
- ๐ feat: Real-Time Context Window & Token Usage Tracking by @danny-avila in #13670
- ๐ฌ feat: Report Tool Results Per Call via
onResultChannel by @danny-avila in #13698 - โ feat: Show Message Timestamps on Hover by @danny-avila in #13709
- ๐ฌ feat: Conversation Starters for Model Specs by @danny-avila in #13710
- ๐พ feat: Persist Context Breakdown & Branch/Total Usage Cost by @danny-avila in #13734
- ๐ธ feat: Per-Agent Endpoint Token Config in Multi-Endpoint Billing by @danny-avila in #13738
- ๐ช feat: Context Gauge UX, Hover Snapshot, Click Breakdown, Currency, Cost-On-By-Default by @danny-avila in #13739
- ๐จ feat: Custom Headers on Built-in Provider Endpoints by @danny-avila in #13742
๐ ฐ๏ธ feat: Native Anthropic Provider for Custom Endpoints by @danny-avila in #13748
๐ Fixes
- ๐ญ fix: Preserve Custom Endpoint Reasoning Params by @danny-avila in #13447
- ๐ช fix: Apply Isolation Plugin to Distinct Queries by @danny-avila in #13457
- ๐ช fix: Preserve Grouped Tool Expansion During Streaming by @danny-avila in #13462
- ๐ fix: Gate RUM Proxy Route on the RUM_ENABLED Flag by @upman in #13475
- ๐ท๏ธ fix: Prevent Bedrock Cache Tokens from Inflating Completion Count by @joohan-lee in #13468
- ๐ชค fix: Reload Messages When Reopening a Chat from a Non-Chat Route by @danny-avila in #13501
- ๐ฆ fix: Gate Chat Starts During Readiness by @danny-avila in #13502
- ๐งท fix: Bind Agent File Context to Current Turn by @danny-avila in #13506
- ๐ผ fix: Harden Shared-Link Message Sanitization with an Allowlist by @danny-avila in #13510
- ๐ผ๏ธ fix: Upgrade Framer Motion for Vite 8 Compatibility by @dlew in #13512
- ๐ชก fix: Handle Missing Skill File Upsert Metadata by @danny-avila in #13520
- ๐ป fix: Clear Project-Scoped Landing When the Selected Project Is Deleted by @danny-avila in #13525
- ๐งญ fix: Restore Empty Skill Allowlist Catalog by @danny-avila in #13526
- ๐ชฆ fix: Add Durable MCP Config Tombstones by @danny-avila in #13534
- ๐ fix: Handle Multiple Concurrent MCP OAuth Login Prompts by @janluedemann-esome in #13200
- ๐ fix: Reuse MCP OAuth Authorization URL by @danny-avila in #13532
- ๐ชช fix: Filter ACL Principal Details by @danny-avila in [#13524](https://github.com/danny-avila/LibreChat/...
v0.8.6
https://www.librechat.ai/changelog/v0.8.6
๐๏ธ Highlights
- Agent Skills & Subagents
- Agent Skills package reusable instructions, references, scripts, assets, and permissions into portable capabilities that agents can invoke automatically or on request.
- Subagents let agents delegate specialized work to other agents while preserving upload, user, and MCP context with recursion and graph limits.
- Code Execution & Artifacts
- Text, source-code, DOCX, CSV, XLSX, and PPTX artifacts can render inline or in the side panel with richer previews.
- Artifact editing, Gemini PDF media blocks, model-spec icons, and persistent agent resource files received reliability fixes.
- Observability & Operations
- Added Prometheus metrics, backend OpenTelemetry tracing, SSE lifecycle tracing, HyperDX browser RUM, structured logging context, explicit readiness endpoints, file-log controls, and graceful HTTP shutdown.
- MCP, Auth, and Admin Hardening
- Improved MCP remote proxy support, OAuth audience handling, JWT expiry fallback, optional OpenID/MCP client-secret flows, multi-audience OpenID validation, trusted registration overrides, admin-panel SSO redirects, streamable HTTP limits, and MCP permission enforcement for agent tools.
- Provider and Model Updates
- Added Claude Opus 4.8 support, Gemini 3.5 Flash support, Gemini tool combinations, Gemma 4 thinking-level support, model-aware Google/Gemini max output tokens, Bedrock AWS profile/API key support, and Bedrock guardrail handling.
- Reliability and UI Polish
- Hardened retention semantics, user/config response sanitization, configured rich-text rendering, Redis/in-memory generation cleanup, all-data retention for agent files, MCP title rendering, sidebar/chat race handling, hover actions, shared links, generated file chips, and balance script config loading.
Click to expand Release Notes
For detailed changes in the release candidate, see:
Changes Since v0.8.6-rc1
โจ Features
- ๐ง feat: Support Guardrail Config Option
streamProcessingModeby @dlew in #12815 - ๐ feat: Add Prometheus Metrics Endpoint + AWS Credential Providers by @danny-avila in #13111
- ๐งพ feat: Add Structured Logging Context by @danny-avila in #13110
- ๐ก๏ธ feat: Bedrock Guardrail Config Environment Variable Resolution by @entropic489 in #11717
- ๐ก feat: Add Backend OpenTelemetry Tracing by @danny-avila in #12909
- ๐ท๏ธ feat: Hide Model Spec Badge Rows by @danny-avila in #13124
- ๐ feat: Enable Eager Execution of Tool Calls by @danny-avila in #13192
- ๐ช feat: Add Session Cookie Secure Override by @danny-avila in #13189
- ๐ช feat: Graceful HTTP shutdown on SIGTERM/SIGINT by @pjhampton in #13211
- ๐ฉบ feat: Add Explicit Readiness Endpoints by @danny-avila in #13212
- ๐๏ธ feat: Allow Disabling File Log Transports by @danny-avila in #13215
- โก feat: Add Gemini 3.5 Flash Support by @danny-avila in #13231
- ๐ฃ๏ธ feat: Add MCP Remote Proxy Support by @danny-avila in #13076
- โ๏ธ feat: Add Operational Prometheus Metrics by @danny-avila in #13265
- ๐ฒ๏ธ feat: Trace SSE Stream Lifecycle with OTel by @upman in #13266
- ๐ feat: Add Explicit new Skill Route from Agent Builder by @sand116 in #13119
- ๐ feat: Built-in Build Metadata for Support Triage by @danny-avila in #12756
- ๐ feat: Add AWS Profile Support for Bedrock Credentials by @iElsha in #10504
- ๐ช feat: Add AWS Bedrock API key support by @dustinhealy in #8690
- ๐ฐ๏ธ feat: Support Gemini Tool Combinations by @danny-avila in #13273
- ๐ง feat: Add Claude Opus 4.8 Support by @danny-avila in #13380
- โ feat: Model-Aware Max Output Tokens for Google/Gemini by @danny-avila in #13390
- ๐ก feat: Add Configurable HyperDX Browser Real User Monitoring by @upman in #13287
- ๐ชช feat: Accept Multiple OpenID JWT Audiences by @danny-avila in #13404
- ๐ชช feat: MCP OAuth - Support
audienceparameter for Auth0/Cognito-style providers by @Freudator86 in #13402 - ๐ feat: Extend
thinkingLevelSupport to Gemma 4 Models by @King-of-Infinite-Space in #13088
๐ Fixes
- ๐งฐ fix: Scope MCP Registry Initialization To Config Fingerprints by @danny-avila in #13115
- ๐งน fix: Reset Redis Reorder State After Last Unsubscribe by @sand116 in #13117
- ๐ท๏ธ fix: Harden Helm Chart Tag Parsing by @danny-avila in #13123
- ๐ก๏ธ fix: Sanitize Agent List Skill Scope by @danny-avila in #13122
- ๐๏ธ fix: Protect Model Spec Instructions by @danny-avila in #13125
- ๐ชต fix: Restore Winston Format Factory Shape In Test Mocks by @JorgeCosta87 in #13139
- ๐ก๏ธ fix: Escape People Picker Search Regex by @danny-avila in #13169
- ๐ค fix: Honor OPENID_REUSE_TOKENS in Admin OAuth Exchange by @jangByeongHui in #13154
- ๐ชช fix: Resolve Group-Scoped Config Overrides by @danny-avila in #13176
- ๐งฝ fix: Strip Admin OAuth Redirect Params by @danny-avila in #13181
- ๐ fix: Prefer LibreChat Web Search over Anthropic's when Both Selected by @danny-avila in #13166
- ๐๏ธ fix: Scope Handoff Agent Context Docs by @danny-avila in #13167
- ๐ชช fix: Scope Message Conversation Access by @danny-avila in #13183
- ๐งญ fix: Reduce MCP Registry ACL Lookups by @danny-avila in #13195
- โด๏ธ fix: Stop Double-Wrapping configYamlContent in Helm ConfigMap by @vdittgen in #13198
- ๐ก fix: Handle Pre-Session 406 for Optional SSE MCP Stream by @pnancarrow in #13202
- ๐งฏ fix: Harden Data Retention Semantics by @danny-avila in #13049
- ๐ชช fix: Add Admin Panel SSO URL Config by @danny-avila in #13220
- ๐ฆฃ fix: Response Size Limits for Streamable HTTP MCP Responses by @danny-avila in #13219
- ๐ fix: Improve OpenID Lookup Planning by @danny-avila in #13229
- ๐งฉ fix: Support DocumentDB Prompt Group Lookup by @danny-avila in #13232
- ๐ก fix: Respect Custom Endpoint Stream Us...
v0.8.6-rc1
๐๏ธ Highlights
- ๐ Agent Skills
- Agent Skills let you package reusable agent behavior into
SKILL.mdbundles: instructions, reference files, scripts, assets, and tool permissions can travel together as a capability an agent can invoke automatically, on request, or always-on. - Skills can be imported, size-limited, shared like agents and MCP servers, and scoped per agent, making it easier to build repeatable workflows without baking every instruction into the agent prompt.
- Agent Skills let you package reusable agent behavior into
- ๐ฅ Subagents
- Subagents let agents call other agents as tools, so a primary agent can delegate specialized work, run independent passes in parallel, or even call a configured copy of itself for fan-out style problem solving.
- Subagent runs preserve the surrounding upload, user, and MCP context while enforcing recursion and graph limits, which keeps multi-agent execution powerful without making it unbounded.
- ๐ฅ๏ธ Code Execution & Artifacts (pending OSS Code Interpreter release)
- Text, source-code, DOCX, CSV, XLSX, and PPTX artifacts can render inline or in the side panel with richer previews.
- Nested artifact paths, Unicode filenames, file metadata encoding, and generated-code context filtering received several hardening passes.
- โ๏ธ CloudFront + S3 File Delivery
- New CloudFront file strategy support includes signed cookies, signed downloads, strict signed-access enforcement, region-aware storage keys, and cookie refresh on auth refresh.
- ๐ Security, Auth, and MCP Hardening
- Improved security for MCP OAuth, OpenID/OIDC, artifact routing, permission caching, avatar fetching, and HTML rendering
- ๐ Providers, Search, and Model Support
- Tavily search/scraping, OpenRouter prompt-cache configuration, Vertex AI multi-region endpoints, Anthropic tool-argument streaming, and GPT-5.5 token definitions are now supported.
Click to expand Release Notes
What's Changed
โจ Features
- feat: agent skills by @danny-avila in #12625
- ๐ Auto-render Text-Based Code Execution Artifacts Inline by @danny-avila in #12829
- ๐ชข Enable Tool-Output References for Bash Tool by @danny-avila in #12830
- ๐ช Render Code-Execution Text Artifacts as Side-Panel Artifacts by @danny-avila in #12832
- ๐ช Render Source-Code Artifacts in the Side Panel by @danny-avila in #12854
- ๐ซ Add Support for
noneReranker Type in Web Search Config by @dlew in #12765 - ๐ญ Require Explicit Auto-agent Enablement for Memories by @danny-avila in #12886
- ๐ช Add allowedAddresses Exemption List For SSRF-Guarded Targets by @danny-avila in #12933
- ๐ Add Tavily as Search and Scraper Provider by @yashwanth-alapati in #12581
- ๐ OIDC Bearer Token Authentication for Remote Agent API by @SpectralOne in #12450
- ๐ Rich File Artifact Previews for DOCX, CSV, XLSX, PPTX by @danny-avila in #12934
- ๐ฉ๏ธ CloudFront CDN File Strategy by @AtefBellaaj in #12193
- ๐งต Enable Anthropic Tool Argument Streaming by @danny-avila in #12962
- ๐ Decouple File Attachment Persistence from Preview Rendering by @danny-avila in #12957
- ๐งฎ Add GPT-5.5 Token Definitions by @danny-avila in #12973
- ๐งญ Add Message Navigation Strip & Redesign Scroll-to-Bottom by @berry-13 in #12657
- ๐ Add Signed CloudFront File Downloads by @danny-avila in #12970
- ๐ฅ๏ธ Add Optional Region-aware S3/CloudFront Storage Keys by @danny-avila in #12987
- ๐ Cross-Origin Admin OAuth Refresh by @dustinhealy in #13007
- ๐งญ Add OpenRouter Prompt Cache Setting by @danny-avila in #13029
- ๐ Mint Code API Auth Tokens by @danny-avila in #13028
- ๐ Support Vertex AI Multi-Region Endpoints by @danny-avila in #13044
- ๐ฆ Configure Skill Import Size Limit by @danny-avila in #13073
- ๐ญ Support OpenID Audience On Refresh Grants by @danny-avila in #13077
- ๐ฉ๏ธ Strict CloudFront signed cookie enforcement via
requireSignedAccessby @upman in #13078
๐ Fixes
- ๐จ๏ธ Restore ModelSpec Preset Greeting (and iconURL Fallback) by @danny-avila in #12809
- ๐ก๏ธ Prevent silent crash from unhandled MCP OAuth reconnect rejections by @danny-avila in #12812
- ๐ฑ Inject Code-Tool Files Into Graph Sessions on First Call (+ read_file Sandbox Fallback) by @danny-avila in #12831
- ๐ซ Reject Binary Files in read_file Sandbox Fallback (No More Mojibake) by @danny-avila in #12851
- ๐ง Replace Literal NUL Bytes in handlers.spec Test Fixture + Normalize CRLF by @danny-avila in #12852
- ๐ Preserve Nested Folder Paths for Code-Execution Artifacts by @danny-avila in #12848
- ๐ Skip Inherited / Mark Skill Files Read-Only in Code-Env Pipeline by @danny-avila in #12866
- ๐ Stop Double-Counting Cache Tokens for Gemini/OpenAI in Usage Spend by @danny-avila in #12868
- ๐ก๏ธ Filter
user_providedSentinel in Tool Credential Loading by @Falenos in #12840 - ๐งน Graceful MCP OAuth Revoke Cleanup When Tokens Are Missing by @gaurav0107 in #12825
- ๐ Prevent Repeated Idle Check Triggers for Users With Failed MCP Connections by @darthhexx in #12853
- ๐งโโ๏ธ Preserve Mermaid
foreignObjectHTML in Sanitized SVG by @ethanlaj in #12819 - ๐ฉน Treat
responseCode === 0as Transport Failure, Not Server Error by @derhelge in #12834 - ๐ Preserve Nested Skill Paths in Code-Env Uploads by @danny-avila in #12877
- ๐ฉน Polish code-execution attachment UX by @danny-avila in #12870
- ๐ Follow 307/308 redirects in MCP streamable HTTP transport by @ontl in #12850
- ๐ฅ Resolve Imported-Conversation Default Model From Runtime modelsConfig by @danny-avila in #12885
- ๐ฉน Sync ControlCombobox popover width with trigger after layout changes by @ethanlaj in #12887
- ๐ก๏ธ Handle MCP Tool Cache Lookup Failures by @danny-avila in #12910
- ๐ Stabilize Agent Prompt Cache Prefix by @danny-avila in #12907
- ๐งญ Migrate Anthropic Long Context by @danny-avila in #12911
- ๐งน Clear MCP OAuth Tokens On Revoke by [@danny-avila](https://github.com/danny-avi...