Index / TinyAGI · updated Sep 3, 2026
TinyAGI
TinyAGI/tinyclaw · healthy · rank 16 of 43 by stars
A multi-agent, multi-team orchestrator (formerly TinyClaw) that runs isolated AI agent teams 24/7 with a slick web 'TinyOffice' dashboard. It leans into the 'one person company' fantasy with team chat rooms, task boards, and pixel-office visualizations.
Facts
Repository
Runtime
Posture
Community
Security breakdown
Composite 55 / 100 · how these are scored
higher is safer
higher is safer
higher is safer
higher is safer
higher is riskier
Model access
Read from the repository, not written by a model · 32 files examined
1 provider · custom endpoint
Pinned models
gpt-5.3-codex released 2026-02-24 — 6 months old (from the public model catalogue)
Pin last edited 6 months ago
Evidence
Decision
Why choose TinyAGI over OpenClaw?
Why choose this
- Native multi-team orchestration with agent hand-offs and fan-out
- Polished web UI with task boards, org charts, and live team visualizer
- SQLite-backed queue with retry and dead-letter handling
Tradeoffs
- Experimental stability vs OpenClaw's more mature single-assistant core
- Fewer messaging channels (3 vs OpenClaw's 8+)
- Docker runs as root; sandboxing delegated to external gVisor
Best fit
- Solo founders wanting multiple agent teams with isolated workspaces
- Users who want a visual dashboard (TinyOffice) for agent management
- Teams needing Discord/WhatsApp/Telegram agent channels
Avoid if
- You need production-stable software (flagged experimental)
- You want strong sandboxing without configuring gVisor yourself
- You prefer a single lightweight assistant over a heavy orchestrator
Evidence is solid from README and recent commits showing active development, but there is zero Reddit discussion and minimal independent web coverage, so community sentiment is largely inferred from stars alone.
AI layer reviewed Aug 24, 2026 · how this is written
Star activity
3,612 stars today
Overview
TinyAGI (formerly TinyClaw) is a TypeScript-based multi-agent orchestrator aimed at the 'one person company' use case: you run multiple teams of AI agents, each with isolated workspaces and specialized roles, collaborating via chain execution and fan-out. It supports Anthropic Claude and OpenAI Codex providers (plus any compatible endpoint), persists sessions across restarts, and uses a SQLite queue with atomic transactions, retries, and dead-letter management.
Its standout feature is TinyOffice, a browser-based control plane with a pixel-office scene, team chat rooms, Linear-style task/project management with comments, and a live TUI dashboard. Agents connect to Discord, WhatsApp, and Telegram, and a plugin system allows custom message hooks and event listeners. Recent commits show heavy investment in the office UI, task management, and daemon control-plane features like restart loops and pairing APIs.
Compared to OpenClaw, TinyAGI trades single-assistant maturity and channel breadth for team orchestration and visualization. Security is a mixed bag: it passes IS_SANDBOX env vars and relies on gVisor for container isolation, but the Docker image runs as root and agents invoke CLI tools with broad shell access. It is explicitly flagged experimental, so expect rapid change and occasional breakage.