GAIA — Personal AI Assistant
A proactive personal AI assistant that connects your tools, runs repetitive work in the background, and reaches you on WhatsApp, Telegram, Slack, or Discord when something needs you.
Evidence shows mature supply-chain practice: extensive pinned security overrides in package., a 3-day PyPI cooldown and dependency constraints in pyproject.toml, plus gitleaks/semgrep/pip-audit/trivy in CI (deduction: trivy runs advisory-only with exit-code 0, so scans do not block). Data-flow transparency rests on FAQ claims (OAuth consent, encrypted tokens, no persistence of integration content) — asserted, not demonstrated in provided code, so sensitive_data_handling gets 2. The product is a proactive multi-channel outbound assistant (WhatsApp/Telegram/Slack/Discord, scheduled workflows): large external-effects surface, mitigated but not eliminated by the notification approval gate, hence 2. Deploy rollback workflows (action=rollback with digest/last modes) justify 2.
CI docs describe a four-slice test matrix, 80% coverage gate, wide-event log contract checks, fail-loud policies and a runner watchdog — self-consistency documented, but no actual test code in the sample, conservatively 2. Lockfiles, pnpm overrides and Docker Compose support availability: 2. Failure messaging is strong at the infra level (orphan-image alerts, explicit step failures) but runtime product error handling is not evidenced: 2.
README targets personal-productivity users across cloud/self-host/desktop/mobile with concrete scenario examples — 3. Capability boundaries are partially marked (beta labels) but the autonomy-vs-approval boundary is explained informally — 2. Cron and multi-platform event triggers documented at moderate precision — 2. Environment fit documented (Docker 20.10+, Node 20+, Tailscale) but self-hosting complexity is high — 2.
Information architecture is excellent: TOC, monorepo map, docs site — 3. Install notes are concrete (3-step CLI, prerequisites, status/logs commands) — 3. FAQ is deep and practical — 3. Biggest deduction: licensing. LICENSE.md is PolyForm Noncommercial 1.0.0, which forbids commercial use, directly contradicting the repeated 'open source' claim in the README; registry metadata is NOASSERTION — a claim contradicted by evidence, so license gets 1. Release-please and conventional commits exist but no CHANGELOG in the sample — 2. Contributing guides and active CI show maintenance responsibility — 2.
The product targets a real pain point (cross-tool memory plus proactive workflows) with strong marginal value — 3. Output usability supported by screenshots and examples but unverified by execution — 2. Cost/benefit is transparent about real self-hosting costs (model APIs, VM), yet the many-database, many-queue architecture raises operational cost — 2.
README's CI/deploy claims corroborate against the workflow files themselves (build, rollback, release flows match in detail) — 2. Most assertions trace to docs or specific workflows, but the data-security claims (encrypted tokens, no persistence) point to code not present in the sample — 2. Facts and inferences are mostly separated (beta and coming-soon labels) — 2.
- License is PolyForm Noncommercial 1.0.0, forbidding commercial use; the README's 'open source' framing is misleading — verify licensing before any commercial adoption.
- Proactive multi-channel outbound (WhatsApp/Telegram/Slack/Discord) plus background workflows create a large external-effects surface: audit trigger conditions and the approval gate on outbound notifications before deploying.
- Data-security claims (encrypted tokens, no persistence of integration content) are README assertions only; verify storage and logging paths in code.
- Trivy runs in advisory mode (exit-code 0), so critical/high findings do not block release; no equivalent npm audit gate is evidenced.
- Publisher is unverified; self-hosting wires in many third-party SaaS services (Composio, E2B, Tavily, etc.) — watch supply-chain and outbound data paths.
What does this agent do, and when should you use it?
GAIA is an open-source personal AI assistant (GitHub: theexperiencecompany/gaia) positioned as 'an assistant, not a chatbot.' It connects tools like Gmail, Calendar, Slack, Linear, and Notion, runs multi-step workflows in the background, and proactively contacts the user based on schedules (cron-style) or events (new emails, calendar changes, webhooks). The project is an Nx-managed full-stack monorepo: a Next.js web app, Electron desktop app, React Native mobile app, a FastAPI + LangGraph backend, a voice processing worker, and Discord/Slack/Telegram bots. Users interact via web, desktop, mobile, or messaging platforms, and GAIA maintains cross-tool long-term memory of people, projects, and preferences. It is available as a zero-setup cloud service and is fully self-hostable via an official CLI running everything in Docker; the code is licensed under PolyForm Noncommercial, permitting noncommercial use only.
GAIA connects to Gmail, Calendar, Slack, Linear, Notion, and more through an integration marketplace (/integrations page) using OAuth or API keys, and can attach any MCP server. Concrete capabilities include: (1) ranking and summarizing unread email threads and drafting replies in the user's voice; (2) watching calendar changes and replanning the afternoon against pending todos and goals when a meeting is cancelled; (3) turning meeting transcripts into action items and creating assigned Linear issues; (4) persistently monitoring the inbox (e.g., for investor emails) and pinging via Telegram within 60 seconds; (5) running scheduled workflows such as a Monday 9am meeting-prep briefing per calendar event. Results land as drafts, todos, reminders, or suggested calendar events in the notifications bell and /notifications page, where the user can approve, edit, or dismiss them. Long-term memory is viewable, deletable, and exportable as a list or interactive graph from the Memory settings page.
- Individual knowledge workers: have GAIA summarize 47 unread emails and draft replies for the 3 that need one
- Engineering leads: auto-post a Friday digest of GitHub, Linear, and Slack activity to #eng-updates
- Frequent meeting attendees: generate a prep doc covering everything shipped this sprint before a 1:1
- Sales/investor relations: monitor email from a specific sender and get a Telegram ping within 60 seconds
- Self-hosters: run on your own server via Docker with full data control and no usage caps
- Inbox-zero practitioners: auto-draft follow-ups to emails sent over 3 days ago without a reply
What are this agent's strengths and limitations?
- Genuinely proactive: event-driven triggers (new email, calendar changes, webhooks) plus scheduled workflows mean no per-task prompting
- Multi-channel reach: message GAIA on WhatsApp, Telegram, Slack, or Discord with full context carried across
- Fully self-hostable: data control, no usage caps, and free model swapping, deployed with a single CLI command
- Cross-tool long-term memory: visualized, deletable, exportable; integration content is fetched on demand rather than persisted
- Licensed under PolyForm Noncommercial 1.0.0 — noncommercial use only; commercial deployment requires separate evaluation
- Self-hosting requires running the full stack (PostgreSQL, MongoDB, Redis, ChromaDB, RabbitMQ, etc.) and supplying your own model API keys; model usage costs scale with usage
- Cloud tier does not allow bring-your-own API keys; model selection is gated by subscription plan
- Desktop and mobile apps are marked beta; iMessage integration is not yet available
How do you install or deploy this agent?
Cloud: visit https://heygaia.io directly — no installation required. Self-host: requires Docker (Engine 20.10+, Compose v2+) and Node.js 20+. Install the CLI and initialize:
npm install -g @heygaia/cli
gaia init # choose "Self-Host (Docker)"The wizard checks prerequisites, clones the repo, guides you through environment variables, then builds and starts every service in Docker; open http://localhost:3000 when done. Already cloned? Run gaia setup inside the checkout. Management commands: gaia status (health check), gaia logs, gaia stop / gaia start.
How do you use this agent?
Use the web UI at http://localhost:3000 (or heygaia.io), connect Gmail, Calendar, Slack, Linear, or Notion via Connect on the /integrations page (OAuth consent) or paste API keys. You can also chat directly on WhatsApp (/gaia commands), Telegram (@heygaia_bot), Slack, or Discord via /gaia or @mention. Build scheduled (cron-style) or event-driven workflows on the Workflows page; outputs appear in the header notification bell and /notifications page for approval. Long-term memory is manageable and exportable (PNG/SVG) on the Memory settings page. Self-hosted instances use your own model provider API keys configured via environment variables.