1. Home
  2. Changelog
Release notes · updated June 2026

Odysseus AI Changelog.

Version history, new features, and bug fixes for Odysseus AI — PewDiePie's self-hosted AI workspace. Launched May 2026 and updated weekly. This page tracks everything since the public launch.

Project · pewdiepie-archdaemon/odysseus License · MIT Latest · v0.7.x (June 2026)
61k+GitHub stars
6Weeks since launch
270+Supported models
~weeklyRelease cadence

This changelog is maintained by odysseusai.site based on official release notes in the GitHub repository. For the authoritative source, see github.com/pewdiepie-archdaemon/odysseus/releases.

Filter:
v0.7.x June 2026 (latest)
FeaturePerf
Deep Research 2.0 · MCP stability · Model cookbook v3

The June update cycle focused on Deep Research quality and the MCP integration layer, two of the most-requested areas since launch.

  • Deep Research now synthesizes across up to 20 concurrent web sources with citation tracking per paragraph
  • MCP server connection stability overhauled — reconnects automatically after network drops
  • Model cookbook expanded to 270+ entries, now scored across five hardware tiers (from 8 GB RAM to 128 GB+ workstation)
  • Agent step logging: each autonomous agent action is now visible in a collapsible trace panel
  • Ollama connection retry logic — no more "provider offline" errors on slow first-boot
  • Code interpreter sandbox now persists state across turns in a conversation
Upgrade: git pull && ./start.sh (Linux/Mac) or .\launch-windows.ps1 (Windows). No database migration needed for v0.6.x → v0.7.x.
v0.6.0 Late May 2026
FeatureBreaking
Email & calendar AI tools go stable · Agent v2 architecture

The email and calendar integration layer, previously in experimental preview, shipped stable in v0.6. This is the release that made "AI that does things" a first-class experience in Odysseus AI.

  • Email tools: IMAP read, SMTP draft and send, thread summarization, priority triage queue
  • Calendar tools: iCal read, event creation, availability checking, meeting proposal drafting
  • Agent v2: new planning engine that generates and validates a step-by-step plan before executing, reducing failed agent runs by ~60% in testing
  • Agent memory: agents can now recall context from previous sessions in the same workspace
  • Breaking: Agent API format changed — any community integrations built against v0.5.x agent endpoints need updating
v0.6.0 breaking change: update any custom agent integrations before upgrading from v0.5.x. Chat history and settings are unaffected.
v0.5.x Mid May 2026
FeaturePerf
Sandboxed code interpreter · Docker GPU passthrough · Windows installer
  • Built-in Python code interpreter — sandboxed execution, file I/O, matplotlib charts rendered inline
  • Docker Compose GPU passthrough finalized for NVIDIA (CUDA) and AMD (ROCm via device passthrough)
  • Official Windows PowerShell launcher (launch-windows.ps1) replaces manual venv setup
  • Performance: streaming response latency reduced ~30% on macOS via Metal Ollama path optimization
  • Memory system: conversation memory now persists across browser sessions
  • RAG: document upload accepts PDF, DOCX, TXT, and Markdown
v0.4.x Mid May 2026
FixPerf
Post-launch stability wave · Ollama bridge fixes

After the PewDiePie launch video drove 61k stars in the first week, this update cycle focused entirely on stability under real-world install conditions.

  • Fixed: host.docker.internal resolution on Linux Docker hosts — extra_hosts now injected automatically when running in Docker on Linux
  • Fixed: Windows PATH not picking up Python after Homebrew-style install — launcher now explicitly resolves venv Python path
  • Fixed: macOS AirPlay port 7000 conflict — launcher now defaults to 7860 on macOS and displays a clear message explaining why
  • Fixed: Admin password not printed on re-runs — now consistently displayed on first boot only, with a reset command documented
  • Performance: cold start time reduced from ~45s to ~12s on average hardware
  • Performance: first-token latency on Ollama path reduced ~20% via async connection pooling
v0.3.0 Early May 2026
Feature
Multi-provider API support · Web search · Model cookbook v1
  • OpenAI, Anthropic, OpenRouter, and 30+ OpenAI-compatible endpoints all configurable in Settings → Providers
  • Built-in web search via Brave Search API (SearXNG and Tavily also supported)
  • Model cookbook: first version with 120 models, hardware-tier scoring, and ollama pull one-click commands
  • Conversation memory: opt-in persistent memory that surfaces relevant context across sessions
  • Dark mode: respects system preference, toggleable manually
v0.2.0 Early May 2026
Feature
Autonomous agents v1 · MCP protocol · Docker Compose
  • Agents v1: multi-step task execution with tool use — web search, file read/write, code execution
  • MCP (Model Context Protocol) integration: connect external tools and data sources natively
  • Official Docker Compose file and docker-compose.yml included in the repo
  • Multi-user support: basic admin and user roles, password authentication
  • RAG v1: document upload and retrieval-augmented chat
v0.1.0 May 12, 2026 — public launch
Feature
Initial public release — "Odysseus AI sets sail"

Odysseus AI launched publicly alongside the PewDiePie YouTube video "I Built My Own AI. You Can Too." The project reached 10k GitHub stars within 6 hours and 61k+ within the first week.

  • Core chat workspace with Ollama local model support
  • OpenAI-compatible API backend — connect any OpenAI-compatible provider
  • Clean browser-based UI with conversation management
  • macOS launcher (start-macos.sh), Linux launcher (start.sh)
  • MIT license — fully open source, forever
The original launch README contained an error in the Windows install path — fixed in v0.1.1 three hours after launch.
Version numbering

How Odysseus AI versions work

Odysseus AI follows a loose semantic versioning scheme during the pre-1.0 phase. The v0.x.0 number indicates a significant feature release — a new subsystem, a major API change, or a paradigm shift in how part of the workspace operates. The patch number (v0.x.y) covers bug fixes, performance improvements, and minor additions that don't change existing behavior. Breaking changes are always called out explicitly in the release notes with a migration note.

As of June 2026 the project is at v0.7.x, moving quickly toward a v1.0 target once the core feature surface — chat, agents, email/calendar, RAG, MCP, and multi-user — is considered stable and well-tested. The public GitHub milestones track progress toward each planned version.

To check which version you have installed, run git log --oneline -1 inside the project folder, or look for the version badge displayed in the Odysseus AI workspace footer after login.

Track Odysseus AI updates

The official release notes live on GitHub. This page summarizes them — for real-time notifications, watch the repo or follow the community Discord.

Official releases on GitHub → Open issues
FAQ

Odysseus AI updates FAQ.

How often does Odysseus AI release updates?
Odysseus AI has been releasing updates approximately weekly since the May 2026 public launch. The project is in rapid development — Felix (PewDiePie) and contributors are actively shipping features. Major version bumps (v0.x) arrive every few weeks; patch releases are more frequent. Watch the GitHub repository for real-time notifications.
Do I lose my chat history when I update Odysseus AI?
No — chat history, settings, and API keys are stored in volumes or the local data directory and are not touched by git pull + relaunch. The only exception is a breaking change noted explicitly in the release notes (v0.6.0 broke the agent API format, but left chat data untouched). Always read the release notes before upgrading a production install.
How do I update Odysseus AI to the latest version?
Inside the project folder, run git pull to fetch the latest code, then relaunch: ./start.sh on Linux/macOS, or .\launch-windows.ps1 on Windows. The launcher detects changed Python dependencies and reinstalls them automatically. For Docker installs: docker compose pull followed by docker compose up -d.
What was the first version of Odysseus AI?
v0.1.0 launched publicly on May 12, 2026, alongside PewDiePie's YouTube video "I Built My Own AI. You Can Too." The initial release included core chat, Ollama support, and the macOS and Linux launchers. The Windows launcher, agents, email tools, code interpreter, and the full model cookbook were added in subsequent updates over the following weeks.
Where is the official Odysseus AI changelog?
The authoritative release notes live at github.com/pewdiepie-archdaemon/odysseus/releases. This page is an unofficial summary maintained by odysseusai.site. For production upgrades, always check the official notes for breaking changes before pulling.
Is Odysseus AI still in beta?
Yes — the version numbers (v0.x) reflect active pre-1.0 development. The project is highly functional for daily use but the API surfaces and configuration formats are still evolving. Expect occasional breaking changes between minor versions, always documented in release notes. The project aims for a v1.0 stable release once the core feature set is complete and hardened.
Continue

Your next step.