What is eclaire?

Eclaire is a privacy-focused AI assistant you host yourself. It centralizes your personal information — to-dos, notes, files, images, and saved links — with a local LLM that can chat about your data, run OCR, tag content, and automate workflows. You run it via Docker Compose as a single container or split services, backed by PostgreSQL or SQLite. The web interface is a React PWA with dark mode, works on mobile, and connects to Apple Shortcuts or Android automation tools. A full OpenAI-compatible REST API lets you build custom integrations. The project is in pre-release (v0.6.x) so expect frequent updates and breaking changes.

Should you self-host eclaire?

eclaire scores 50/100 on selfhost.directory (how scoring works) and ranks #40 of 120 in AI Assistant. It has 926 GitHub stars.

It shipped 8 stable releases in the past year. The latest, v0.6.3, came out on 6 March 2026. New versions land roughly every 2 weeks.

Setup is advanced: plan for several services and some ongoing maintenance. An official Docker image is available. It is MIT-licensed and written mostly in Typescript.

Its closest alternative on selfhost.directory is hermes-agent (89/100, 258 015 stars). It scores higher overall, so compare the two before you pick. See eclaire vs hermes-agent side by side.

Features

Admin CLI
AI Assistant with Tool Calling
Bookmark Management
Document Processing & OCR
Flexible Deployment Modes
Hardware Acceleration
Multiple LLM Backend Support
Notes with Rich Text/Markdown
OpenAI-compatible REST API
Photo Management & OCR
Scheduled Automations
Task Management
Unified Data Management

How to configure eclaire

AI_TIMEOUT
ANTHROPIC_API_KEY
API_KEY_HMAC_KEY_V1
BACKEND_URLBackend URL (defaults to FRONTEND_URL in containers, http://127.0.0.1:3001 locally) In Docker, frontend and backend are the same origin — set this to the same value as FRONTEND_URL if you override it.
Backend URL (defaults to FRONTEND_URL in containers, http://127.0.0.1:3001 locally) In Docker, frontend and backend are the same origin — set this to the same value as FRONTEND_URL if you override it.
BETTER_AUTH_SECRETSecurity Secrets (REQUIRED - generated by setup scripts) Generate manually with: openssl rand -hex 32
Security Secrets (REQUIRED - generated by setup scripts) Generate manually with: openssl rand -hex 32
DATABASE_HOSTPostgreSQL connection (optional - defaults work with included postgres container) Host auto-detected: 'postgres' in containers, '127.0.0.1' locally
PostgreSQL connection (optional - defaults work with included postgres container) Host auto-detected: 'postgres' in containers, '127.0.0.1' locally
DATABASE_NAME
DATABASE_PASSWORD
DATABASE_PORT
DATABASE_TYPEpostgresDatabase Configuration Default: postgres (recommended for dev/prod parity) Options: sqlite | pglite | postgres
Database Configuration Default: postgres (recommended for dev/prod parity) Options: sqlite | pglite | postgres
DATABASE_URLOr provide a full connection string (overrides individual settings)
Or provide a full connection string (overrides individual settings)
DATABASE_USER
DOCLING_SERVER_URLExternal Services Docling URL auto-detected: 'http://docling:5001' in containers, 'http://127.0.0.1:5001' locally Override only if your setup differs:
External Services Docling URL auto-detected: 'http://docling:5001' in containers, 'http://127.0.0.1:5001' locally Override only if your setup differs:
FRONTEND_URLPublic-facing URL (for OAuth callbacks, email links, etc.) Set this to the URL users type in their browser. Required for: - Reverse proxy setups (e.g. https://app.example.com) - LAN access (e.g. http://nas.local:3000) For plain localhost, the defaults work out of the box.
Public-facing URL (for OAuth callbacks, email links, etc.) Set this to the URL users type in their browser. Required for: - Reverse proxy setups (e.g. https://app.example.com) - LAN access (e.g. http://nas.local:3000) For plain localhost, the defaults work out of the box.
GITHUB_TOKENGitHub token (for higher rate limits on GitHub bookmark processing)
GitHub token (for higher rate limits on GitHub bookmark processing)
HOST_PORTServer Host port for Docker (the container always listens on 3000 internally)
Server Host port for Docker (the container always listens on 3000 internally)
LLAMA_CPP_BASE_URLAI Configuration Providers and models are configured via the web UI or CLI. Local LLM server URLs are auto-detected based on ECLAIRE_RUNTIME: - Container: http://host.docker.internal:<port>/v1 - Local: http://127.0.0.1:<port>/v1 Override only if your setup differs from defaults:
AI Configuration Providers and models are configured via the web UI or CLI. Local LLM server URLs are auto-detected based on ECLAIRE_RUNTIME: - Container: http://host.docker.internal:<port>/v1 - Local: http://127.0.0.1:<port>/v1 Override only if your setup differs from defaults:
LLAMA_CPP_BASE_URL_2
LM_STUDIO_BASE_URL
LOG_LEVELPerformance Tuning (optional)
Performance Tuning (optional)
MASTER_ENCRYPTION_KEY
MLX_LM_BASE_URL
MLX_VLM_BASE_URL
OLLAMA_BASE_URL
OPENAI_API_KEYCloud provider API keys (referenced by AI provider config in DB)
Cloud provider API keys (referenced by AI provider config in DB)
OPENROUTER_API_KEY
PGLITE_DATA_DIR
PORTPort for local (non-Docker) development
Port for local (non-Docker) development
QUEUE_BACKENDpostgresQueue backend (set by setup script to match database type) Options: sqlite | postgres Note: sqlite queue requires SERVICE_ROLE=all and DATABASE_TYPE=sqlite
Queue backend (set by setup script to match database type) Options: sqlite | postgres Note: sqlite queue requires SERVICE_ROLE=all and DATABASE_TYPE=sqlite
REDDIT_CLIENT_IDReddit API (for Reddit bookmark processing)
Reddit API (for Reddit bookmark processing)
REDDIT_CLIENT_SECRET
SERVICE_ROLEService Mode Default: all (unified mode - API + workers in single process) Options: all | api | worker
Service Mode Default: all (unified mode - API + workers in single process) Options: all | api | worker
SQLITE_DATA_DIRSQLite/PGlite data directories (only if using sqlite/pglite)
SQLite/PGlite data directories (only if using sqlite/pglite)
WORKER_CONCURRENCY
X_BEARER_TOKENX (Twitter) API — Tweet Lookup (app-only, no user OAuth needed) Bearer Token from your X app's "Keys and tokens" page (OAuth 2.0 App-Only) Requires Basic tier ($200/mo) — https://developer.x.com/en/portal/dashboard
X (Twitter) API — Tweet Lookup (app-only, no user OAuth needed) Bearer Token from your X app's "Keys and tokens" page (OAuth 2.0 App-Only) Requires Basic tier ($200/mo) — https://developer.x.com/en/portal/dashboard
X_CLIENT_IDX (Twitter) API — Bookmarks Sync (requires per-user OAuth) Client ID and Secret for OAuth 2.0 with PKCE authorization flow Only needed if you want users to import their X bookmarks
X (Twitter) API — Bookmarks Sync (requires per-user OAuth) Client ID and Secret for OAuth 2.0 with PKCE authorization flow Only needed if you want users to import their X bookmarks
X_CLIENT_SECRET

FAQ

How can I contribute or report issues?

Star the repo, open issues/PRs, and join discussions. The project welcomes bug reports, fixes, new features and integrations.

Can I export or delete my data?

Yes, you retain full ownership and can export or remove your data at any time via the client or APIs.

Is it really open-source?

Yes, Eclaire is fully open-source. The LICENSE file in the GitHub repo provides details.

Is there a hosted version available?

Currently, Eclaire is only available as a self-hosted solution to ensure maximum privacy and control. However, business consulting and managed deployment services are offered for organizations that need assistance.

What are the system requirements?

Eclaire runs on Mac Mini/Studio (M1 or above) and on Linux/Windows with GPUs. The amount of memory needed depends on the AI models selected.

Can I integrate Eclaire with my existing tools?

Yes, Eclaire provides a comprehensive REST API and supports integrations with popular tools. It also works seamlessly with Apple Shortcuts, Android automation, and can be extended with custom integrations.

What AI models does Eclaire support?

Eclaire supports popular open-source models including Qwen, Gemma, Deepseek, Kimi, Llama, and many others. You can download models from Hugging Face or connect it to cloud inference APIs, and easily switch between models or run multiple models simultaneously.

How is Eclaire different from cloud-based AI assistants?

Eclaire runs entirely on your own hardware, ensuring complete privacy and data sovereignty. Your data never leaves your control, and you're not dependent on external services or subject to their terms and privacy policies.