Changelog
Release notes for Auxot Server and Auxot OSS.
Native Gemini integration and admin password reset
Added native Google Gemini API support and admin-initiated password reset via invite links.
Read more →Grok and Gemini provider support with credential and routing improvements
Added Grok (xAI) and Gemini as fully supported cloud LLM providers, alongside fixes for credential resolution, provider routing, and Slack agentic loop continuity.
Read more →Context files, hybrid search, and agent steering
This release ships structured context-file tools with hybrid search, improved agent steering mid-turn, MCP catalog editorial metadata, and several connector and chat reliability fixes.
Read more →ACL pipeline, vector search, and sandbox API additions
This release adds connector ACL pipelines with thread branching and cron routing, vector search for playbook bullets, and expanded sandbox globals, alongside a broad set of fixes for security isolation, credential handling, and UI polish.
Read more →Headless agents and execution context improvements
Introduced headless agent support with full personality and skill configuration, improved execution context panel usability, and fixed several stability issues around cron messaging, GPU worker disconnects, and prompt compaction.
Read more →Worker Reliability Improvements
Improved worker startup resilience and cleaner debug logging for image-heavy workloads.
Read more →Vision pipeline improvements and dynamic worker reconfiguration
Added image support to the CLI worker's read tool and introduced live llama.cpp reconfiguration on policy updates.
Read more →Coordinator reliability and debug improvements
Fixed cloud job provider persistence and continuation loop stalls, with improved thread tool result debug display.
Read more →Embeddings, live CLI streaming, and security hardening
This release introduces ONNX-backed semantic embeddings, early tool-dispatch SSE for the live CLI, playbook thread memory and curator ops, and a series of security fixes across Slack and Discord attachment handling.
Read more →Cross-channel messaging, cron destinations, and admin agent gaps
This release adds cross-channel messaging and cron destinations, closes several Admin Agent capability gaps, tightens tool authorization rules, and improves UI clarity around destructive session actions.
Read more →Build integrity and serving reliability improvements
Improved deployment reliability with build integrity checks, correct SPA cache headers, and a healthcheck for the scratch-image container.
Read more →Scripts page, cron fixes, and tool normalization
Introduced a team-scoped Scripts page with a live code editor, fixed cron scheduling and message routing bugs, and normalized control-plane tools for LLM callers.
Read more →Anonymous access overhaul, trigger policy redesign, and chat reliability
This release overhauls anonymous access and tools-as-code paths, redesigns agent trigger policies as optional context deltas, improves chat reliability with fail-fast error handling, and includes several coordinator, UI, and server fixes.
Read more →HTTP Streamable MCP Client
Added support for HTTP Streamable transport in the MCP client, enabling connections to remote MCP servers over HTTP.
Read more →HTTP MCP, user invites, and provider resilience
This release adds HTTP MCP support, a pending user invite flow, improved provider failover under rate limits, and multimodal tool result handling.
Read more →Playwright browser tooling, bash tool improvements, and browser stability fixes
Adds Playwright MCP browser tooling for agents, improves the bash tool output format, removes MLX/vllm-mlx integration, and ships a series of browser worker stability and transport fixes.
Read more →Coordinator reliability, browser fixes, and design system overhaul
This release improves coordinator cancel/loop detection, fixes browser tool schemas, expands sandbox capabilities, and delivers a comprehensive design system and navigation overhaul.
Read more →Qwen3 model support and per-job model routing
Added Qwen3.6-35B-A3B GGUF to the model registry and fixed CLI workers to correctly use the model specified in server-assigned jobs.
Read more →Tools-by-code, configurable provider fallback, and unified agent scheduling
Introduces tools-by-code with team-scoped scripts and ranked discovery, a configurable provider fallback graph, and unified cron/defer scheduling with improved agent-facing UX.
Read more →File Upload Support and Worker Reliability
Adds worker-side file upload handling with automatic cleanup, embeds built-in Auxot skills, and fixes several worker and path-handling bugs.
Read more →File attachments, multi-model support, and resilient job infrastructure
v0.1.25 brings file attachment support in agent chat, multi-model provider selection for Claude Code, Discord/Slack pass() tool with reprompt loop, prompt cache token metrics, and sweeping improvements to job queue resilience and cloud inference stability.
Read more →Rate-Limit Detection, Job Retry, and Tool-Continuation Fix
The Claude CLI worker now detects provider rate limits and over-quota conditions, reports a usage snapshot to the server for dashboard visibility, and automatically retries jobs after backoff. A tool-continuation pipe bug is also fixed.
Read more →Live-MCP CLI Worker, Gemma 4, and Context Truncation
Live-MCP tool continuation for the Claude CLI worker, Gemma 4 model support, and automatic truncation of large bash/read outputs with recall markers.
Read more →Agentic Apps, Job Retry Resilience, and Rate-Limit Visibility
Introduced an Agentic Apps page for managing agent-backed applications, live provider queue-depth SSE, automatic job retry on transient provider failures, rate-limit visibility on the dashboard, credential editing for team admins, and a simplified activity feed.
Read more →Workflow Reliability, Context Recall, and Debug Visibility
Automatic recovery of in-flight work items after server restarts, context-aware pre-compaction with tool and message recall, full agentic loop reliability fixes, and greatly improved LLM debug visibility.
Read more →Slack Integration, Feature Flags, and UI Overhaul
Added full Slack integration with Socket Mode, a unified feature flag system, OpenAI/Anthropic API image passthrough, agentic-app grouping in system health, and a comprehensive mobile and design system overhaul.
Read more →Multimodal Vision Support
Added end-to-end multimodal (vision) support across the CLI worker and Llama executor.
Read more →Helm chart image pull policy update
Updated the Helm chart to always pull the latest images on deployment.
Read more →Mobile optimization and image upload improvements
This release brings comprehensive mobile and tablet layout improvements across the UI, client-side image resizing for multimodal conversations, and Helm chart updates for more robust agent home directory handling.
Read more →Claude CLI stability and session handling fixes
Improved reliability of Claude CLI integration with better session ID normalization, stale file cleanup, and error propagation.
Read more →Helm chart multi-instance support and Claude Code OAuth
Helm chart 0.3.0 adds multi-instance configuration maps and Claude Code OAuth environment support.
Read more →Discord bot stability improvements
Improved Discord bot reconnect reliability and added optimistic UI status updates.
Read more →Explorer jobs view, context files v2, and reasoning improvements
This release overhauls the Explorer around a persistent jobs view, ships context files v2 with indexed mode and search tools, and improves reasoning token streaming and Discord integration lifecycle management.
Read more →Claude OAuth secrets support
Added support for Claude OAuth secrets configuration.
Read more →Rollout reliability fix
Fixed an issue with the rollout method that could cause incorrect deployment behavior.
Read more →Helm chart enhancements and tools deployment support
Expanded Helm chart configurability with optional Namespace creation, external Secret references, worker ROUTER_URL support, and a dedicated tools Deployment.
Read more →Streaming Reliability Fixes
Resolved issues with streaming output in the agent worker pipeline.
Read more →Teams V2, onboarding wizard, and chat improvements
Introduced Teams V2 with invites and member roles, a guided onboarding wizard, team-scoped agents, and a range of chat and UX improvements.
Read more →Agent Tooling, Docker Improvements, and Connection Stability
Added preliminary agent attach support, improved Docker and worker bootstrapping, and fixed reconnect backoff behavior in the agent worker.
Read more →Agent Worker Refactor and Claude CLI Improvements
Refactored the agent worker, updated Claude CLI integration, and improved protocol message handling.
Read more →Licensing tiers, team management, and Discord bot
Introduced licensing tiers, team invite and default settings, a Discord bot integration, and a refactored coordinator with improved job and thread context handling.
Read more →Worker reliability improvements
Added comprehensive worker test coverage and resolved several production bugs uncovered during testing.
Read more →WebSocket stability and test coverage improvements
Improved agent WebSocket reliability by fixing stale connection handling and expanded test coverage for routing infrastructure.
Read more →WebSocket stability and agent heartbeat
Improved connection reliability for agents with concurrent WebSocket safety and heartbeat support.
Read more →Agent Loop URL Fix & Onboarding Docs
Fixed agent loop URL construction and added onboarding documentation.
Read more →Worker image CA certificate fix
Fixed TLS connectivity in the auxot-worker Docker image by installing CA certificates.
Read more →Unified WebSocket endpoint and router URL normalization
Consolidated WebSocket connectivity under a single endpoint with automatic router URL normalization.
Read more →WebSocket agent endpoint restoration
Restored the /ws/agent endpoint as a full WebSocket connection and fixed heartbeat handling in the tools handler.
Read more →WebSocket agent connection fix
Fixed a routing issue that prevented WebSocket agent connections from being established correctly.
Read more →Redis-backed tools queue and stale job cleanup
Introduced a Redis-backed tools queue with automatic cleanup of stale agent continuations.
Read more →Onboarding and Redis connection improvements
Added onboarding experience with documentation links and improved Redis connection pool configuration.
Read more →Unified WebSocket agent routing
Agent workers now route through a single unified WebSocket endpoint, simplifying connectivity and deployment configuration.
Read more →Worker Reconnect Stability
Improved tools worker reliability by fixing reconnect backoff behavior.
Read more →Docker build fix for npm bin symlinks
Fixed a Docker packaging issue that caused npm binary symlinks to be missing in the final image.
Read more →Agent Node Naming Fix
Corrected the built-in node user name from `node` to `agent` in the agent Dockerfile.
Read more →Official Docker Images for Worker and Agent
Auxot now ships official Docker images for auxot-worker and auxot-agent, making containerized deployments easier.
Read more →Agent Tool Call Reporting Fix
Fixed agent tool call reporting to ensure tool invocations are correctly tracked and surfaced.
Read more →Agent NPM Package
Published the Auxot agent as an npm package for easier installation and distribution.
Read more →Streaming reliability and UX polish
Improved streaming stability and applied UX review updates.
Read more →Licensing and tooling fixes
Improved agent licensing support and resolved initial tooling and audit log issues.
Read more →Claude Code CLI Worker & Daemon Management
Introduced a Claude Code CLI worker, daemon management subcommands, and npm package publishing support.
Read more →Initial release
First release of Auxot Server, featuring multi-model agent orchestration, team management, licensing, and a full-featured chat and settings UI.
Read more →Vision capabilities and default endpoint
Workers now advertise vision support when required by policy and default to the Auxot GPU endpoint.
Read more →Vision, Image Generation, and Model Registry 1.1.2
Added multimodal vision and image generation support, upgraded the model registry to 1.1.2 with Qwen3.5-35B-A3B as the new default model.
Read more →Tools Worker
Introduces the tools worker with web tool support and improved reconnect reliability.
Read more →Improved HTTP Error Logging
Unhandled HTTP requests now log method, path, and remote address for easier debugging.
Read more →Thinking Support and Flexible API Routing
Added support for model thinking/reasoning and tool generation, with more resilient API path handling for OpenAI and Anthropic endpoints.
Read more →Tolerant API path routing
Added middleware to gracefully handle malformed API paths with doubled version segments.
Read more →KV Cache Usage Reporting
Added KV cache hit count to usage reporting for better token cost visibility.
Read more →Usage metrics in streams, README overhaul
Streaming responses now include token usage data, and the README has been rewritten with clearer setup instructions.
Read more →Parallel Dial-Down on OOM
Workers now reduce parallel execution under memory pressure, with improved documentation for base URLs, CLI commands, and deployment.
Read more →Qwen3-Coder default model and context visibility
Switched the default model to Qwen3-Coder-30B-A3B and surfaced context length in model listings.
Read more →Initial Release
First open-source release of the Auxot agent orchestration platform, including the router, worker, and CLI foundation.
Read more →