Eclaire
A self-hosted assistant that organizes and works across your private tasks, notes, files, photos, and bookmarks.
- Source repo
- eclaire-labs/eclaire
- Stars
- ★ 927
- Last updated
- 5mo ago
- License
- MIT
- Primary language
- TypeScript
- FA score
- 57/100 · Major gaps
At a glance
- How it runs
- Works with
- Universal · cross-platformOpenAI API
- Cost
- Free, no paid service needed
- Setup effort
- High · needs real infrastructure
- You'll need
- Typical use
- A privacy-conscious individual can keep bank or health documents on local infrastructure and search or summarize them with a local model.
- Not a fit if
- Users who want a turnkey product without running their own local LLM server
- Teams that need to expose it directly to the public internet with hardened security
- Teams that cannot tolerate frequent breaking changes in a pre-release project
- Source review
- 57/100 · Major gaps
What does this agent do, and when should you use it?
Eclaire is a local-first, self-hosted AI assistant for personal tasks, notes, documents, photos, and bookmarks. Its architecture combines a Vite/React 19 frontend, Node.js/Hono REST API, background workers, a data layer, and local model services; the components can run together in one container by default. The assistant chats over content context and can search data, open content, resolve tasks, add comments, and create notes. PostgreSQL or SQLite provide persistence, while the job queue can use a database or Redis; Docling handles document processing. It exposes OpenAI-compatible REST endpoints authenticated with session tokens or API keys, but remains pre-release and is not intended for direct public-internet exposure.
Eclaire ingests uploaded files, images, tasks, plain-text or Markdown notes, and bookmarks. For bookmarks, it fetches pages and creates PDF, readable, and LLM-friendly versions, with dedicated GitHub and Reddit API and metadata handling. Its workers process documents and images for OCR, content understanding, classification, and search-related workflows across listed Office, PDF, web, and image formats. The chat assistant streams responses grounded in that content and can use tools to search data, open content, resolve tasks, add comments, or create notes. It works with llama.cpp, vLLM, mlx-lm/mlx-vlm, LM Studio, Ollama, and other OpenAI-compatible model backends; ./eclaire model list lists models.
- A privacy-conscious individual can keep bank or health documents on local infrastructure and search or summarize them with a local model.
- A read-later user can save web bookmarks and receive readable and PDF archive versions.
- A Mac user can dictate a note with an Apple Watch, then connect that note to existing tasks and documents.
- Someone with a backlog of scans and photos can run OCR, categorize them, and search for information inside them.
- A person managing personal work can create tasks and have the assistant add comments, resolve tasks, or write results to separate documents.
- A user who needs mobile or messaging access can interact through the installable PWA, iOS Shortcuts, Android automation, or Telegram.
How do you install or deploy this agent?
Install Docker, Docker Compose, and a local LLM server; the README recommends llama.cpp. Run:
mkdir eclaire && cd eclaire
curl -fsSL https://raw.githubusercontent.com/eclaire-labs/eclaire/main/setup.sh | shThe script downloads configuration, generates secrets, and initializes PostgreSQL. Start your model server, run docker compose up -d, then open http://localhost:3000 to create an admin account and configure an AI provider and models; ./eclaire onboard provides a CLI onboarding path. Source development additionally requires Node.js ≥24 with corepack enabled and pnpm.
How do you use this agent?
Start the local model server, then run docker compose up -d. Complete onboarding at http://localhost:3000, add or upload tasks, notes, documents, photos, and bookmarks, and ask the assistant questions over that content. Configure providers, models, and the active selection in the admin UI or CLI; use ./eclaire model list to inspect models. For programmatic access, use the OpenAI-compatible REST endpoints with either session tokens or API keys.
What are this agent's strengths and limitations?
- It places tasks, notes, documents, images, and bookmarks in one data space that the assistant can search and use as chat context.
- It supports a unified single-container deployment as well as separate services, with PostgreSQL or SQLite and database-backed or Redis queues.
- It supports several local or OpenAI-compatible model backends, including llama.cpp, MLX, vLLM, LM Studio, and Ollama.
- Bookmark handling produces PDF, readable, and LLM-friendly copies, with specialized GitHub and Reddit handling.
- The project is explicitly pre-release, with expected breaking changes; operators are told to back up data and review release notes before upgrades.
- Operation requires Docker, Docker Compose, and an independently configured local LLM server.
- It is explicitly not hardened for direct public exposure, so remote access needs additional protection such as a VPN, tunnel, or authenticated reverse proxy.
- Some acceleration paths depend on the available hardware and runtime, such as Apple MLX or NVIDIA CUDA.
How does this agent compare with similar options?
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| Eclaire This agent | 57 · Major gaps | Self-hosted serviceFree | ★ 927 | 5mo ago | TypeScript | OpenAI API |
| Talorys | 83 · Good | — | ★ 1.5k | 2d ago | TypeScript | — |
| GAIA — Personal AI Assistant | 69 · Some gaps | Hosted serviceFreemium | ★ 325 | today | Python | — |
| SiYuan | 40 · Major gaps | Desktop appFreemium | ★ 47k | today | TypeScript | — |
How does FollowAgents rate this agent?
Why each dimension lost points
The evidence shows that the settings guide tells the assistant to check current profile values and confirm before changing them; the media guide also requires confirmation for imports and deletion. The sample gives limited detail on permission boundaries, data flows, and reversal for API, MCP, and external-service actions. The README describes local models and storage but also lists remote model, audio, GitHub, and Reddit integrations without fully explaining when data leaves the machine or how sensitive data is handled, so these scores remain low. The repository provides a vulnerability reporting route, but publisher identity is unverified and earns no credit.
The startup entry logs unhandled rejections and uncaught exceptions before exiting, and configuration validation is described as a startup check. The README explicitly says the project is pre-release and may have breaking changes. Multiple model, database, and queue options provide flexibility, but the sample does not establish availability guarantees or complete recovery behavior, so reliability scores are moderate or low.
The product guide covers notes, bookmarks, documents, media, photos, and tasks; the README adds personal knowledge management, imports, search, automation, and multiple channels, giving strong audience and scenario coverage. The admin guide explains provider, model, and MCP configuration, though MCP security boundaries, capability limits, and precise feature triggers are not fully specified in the sample. Docker, self-hosting, SQLite/Postgres, local models, and cross-platform claims show environment flexibility, while deployments still require Docker and database or model services.
The README separates installation, model selection, architecture, development, and security information and includes setup examples; the sampled skills have named metadata and topic tags. The MIT license text is complete. The pre-release warning, release link, and changelog reference support version and limitation guidance, but the security policy describes volunteer best-effort maintenance with no SLA and support only for main. The changelog body is not included in the packet, so its details cannot be assessed.
The product combines organization, search, summarization, and automation across several personal data types, and the task, media, and settings guides provide actionable user guidance. It has clear value for self-hosted personal knowledge management. However, operation requires model and container services and may require document-processing tools; the sample does not quantify benefits, performance, or maintenance burden, so cost-benefit receives a low score.
README installation, architecture, and security claims can be partly cross-checked against package.json, the startup entry, the security policy, and sampled skills. The CI workflow lists lint, format, typecheck, build, and test jobs, and the database samples contain assertion-based cases. The packet is partial and contains no execution results, so successful runs cannot be verified; many broad README capability claims are also not corroborated by the provided code. Documentation claims and code facts are not always clearly separated.
- The README warns against exposing the service directly to the public internet and says it is not hardened for that exposure; configure an additional private network, tunnel, or authenticated reverse proxy as recommended before deployment.
- The project is pre-release and upgrades may introduce breaking changes; the README recommends regular backups and reviewing release notes.
- The sample lists remote models, audio services, MCP, messaging channels, and URL media imports without fully showing their data flows and permission controls; check the actual configuration before processing sensitive information.
Maintain this project? Add the FARS badge to your README
[](https://followagents.com/en/agents/eclaire)<a href="https://followagents.com/en/agents/eclaire"><img src="https://followagents.com/badge/eclaire.svg" alt="FARS review"></a>Recent releases
Latest entries from the project's GitHub Releases. Notes are quoted from upstream as written, not rewritten or tested by FollowAgents.
-
Release notes excerpt
Bug Fixes - workers: add tiered extraction with lightweight HTTP fetch before browser - backend: include missing thumbnailUrl and faviconUrl in bookmark list responses - auth: harden Better Auth integration — security fixes, fail-fast init, dead code removal - frontend: fix SSE connection leak on page refresh - frontend: improve SSE streaming robustness and eliminate unnecessary re-renders - frontend: fix re-render issues, remove debug logging, and improve data fetching - frontend: remove hardcoded manifest link from index.html …View on GitHub ↗ -
Release notes excerpt
Security - deps: upgrade hono 4.11.4 → 4.12.2 for multiple CVEs (XSS, cache bypass, IP spoofing, timing attack) - deps: upgrade axios 1.13.2 → 1.13.5 for GHSA-43fc-jf86-j433 (DoS via __proto__ in mergeConfig) - deps: upgrade node-gyp 12.1.0 → 12.2.0 and refresh transitive deps (tar, minimatch, ajv, qs, brace-expansion) - deps: resolve seroval and diff vulnerabilities via clean reinstall - frontend: fix XSS vulnerability from unsanitized dangerouslySetInnerHTML uncovered during lint audit CI/CD …View on GitHub ↗ -
Release notes excerpt
Security - backend: upgrade hono to 4.11.4 for CVE-2026-22817 (JWT Algorithm Confusion)View on GitHub ↗
FAQ
Does Eclaire require a hosted model?
Can I expose it directly to the public internet?
What must I do after first startup?
./eclaire onboard.What content can it process?
Thanks — we use this to improve the page.