Key takeaways
- 3,574 GitHub stars as of June 11, 2026 — up only 67 from this site's April snapshot, with no commits pushed since March 30, 2026
- Despite the "agent teams orchestrator for One Person Company" branding, the project is a multi-agent personal assistant — it began life as a "tiny wrapper of Claude Code that acts as your 24/7 personal assistant" — and fits the personal-agents category better than company platforms
- Fully local and lightweight by design: a TypeScript daemon with a SQLite store and isolated agent workspaces, installed with one curl command, MIT licensed, no cloud account required
FAQ
What is TinyAGI?
TinyAGI is an open-source, self-hosted platform for running small teams of AI agents (coder, writer, reviewer, researcher) that collaborate through message queues and shared chat rooms, aimed at solo operators.
How much does TinyAGI cost?
Free — MIT licensed and self-hosted. Real costs are the API fees of whatever model provider you connect (Claude, Codex, or compatible endpoints).
What models does TinyAGI run on?
Anthropic Claude (Opus, Sonnet), OpenAI Codex, and any custom OpenAI- or Anthropic-compatible endpoint.
How is TinyAGI different from Paperclip?
Paperclip models agents as employees in a governed org chart with budgets and approval gates; TinyAGI is a lightweight team of 3-5 personal helper agents with chat rooms and a kanban board — no governance layer, budgets, or formal hierarchy.
Executive Summary
TinyAGI (formerly TinyClaw) is an MIT-licensed, self-hosted orchestrator for small teams of AI agents, self-described as "the agent teams orchestrator for One Person Company."[1] It runs locally as a daemon, routes messages between a handful of specialized agents — @coder, @writer, @reviewer, @researcher — through chat rooms and a message queue, and exposes a web UI and CLI.[2] The project began in February 2026 as a "tiny wrapper of Claude Code that acts as your 24/7 personal assistant" and grew into a multi-team variant before rebranding mid-March.[3][4]
Status note: the repository is not archived, but development has stalled — the last push was March 30, 2026, and the latest release (v0.0.20) shipped March 26, 2026. Stars stand at 3,574 as of June 11, 2026, up just 67 from this site's April 2026 snapshot (3,507).[1][4] Category fit is also in question: despite the "company" branding, what TinyAGI actually ships — a small crew of personal helper agents with no budgets, governance, or org chart — places it closer to personal-agents platforms than to agent company platforms like Paperclip.
| Attribute | Detail |
|---|---|
| Creator | Jian Liao (@jlia0 / @jianxliao)[5] |
| Founded | February 9, 2026 (as TinyClaw)[1] |
| Funding | None disclosed |
| GitHub Stars | 3,574 (as of June 11, 2026)[1] |
| License | MIT[1] |
| Last activity | Pushed March 30, 2026; v0.0.20 released March 26, 2026[1][4] |
Product Overview
TinyAGI's tagline is "Multi-agent, Multi-team, Multi-channel, 24/7 AI assistant." You install it with one curl command, it starts a local daemon, auto-creates a default team, and opens a browser UI. Agents live in isolated workspaces, mention each other to hand off work (@coder fixes, @reviewer checks), and fan out in parallel through an actor-model message queue.[2][5]
There are no fixed organizational roles. Teams have a designated leader agent, persistent chat rooms, and a task kanban board with role-based assignment — but agents are user-defined helpers, not employees in a hierarchy.[2]
Key Capabilities
| Capability | Description |
|---|---|
| Multi-agent teams | Several specialized agents per team, collaborating via @-mentions and message queues[2] |
| Multi-team isolation | Multiple teams run simultaneously with isolated workspaces[2] |
| Leader agents | Each team has a designated leader for coordination[2] |
| Task kanban | Board with role-based task assignment[2] |
| Persistent chat rooms | Ongoing conversations across web and CLI surfaces[2] |
| Office control plane | Web dashboard introduced in v0.0.20[4] |
Product Surfaces
| Surface | Description | Availability |
|---|---|---|
| Web UI | Browser dashboard, opens on launch | Self-hosted[2] |
| CLI | tinyagi command, TypeScript rewrite in v0.0.20 | Self-hosted[4] |
| Docker | docker compose up -d | Self-hosted[2] |
Technical Architecture
TinyAGI runs entirely locally on macOS, Linux, or Windows (WSL2) — as a background daemon or Docker container. State lives in a SQLite database with agent workspaces under ~/tinyagi-workspace/.[2]
curl -fsSL https://raw.githubusercontent.com/TinyAGI/tinyagi/main/scripts/install.sh | bash
tinyagi
The v0.0.15 release added the one-line install and auto-migration from ~/.tinyclaw to ~/.tinyagi, completing the rebrand.[4]
Key Technical Details
| Detail | Value |
|---|---|
| Deployment | Local daemon or Docker; no cloud account[2] |
| Model(s) | Anthropic Claude (Opus, Sonnet), OpenAI Codex, custom OpenAI/Anthropic-compatible endpoints[2] |
| Language | TypeScript[1] |
| Storage | SQLite + per-agent workspace directories[2] |
| Open Source | Yes — MIT, on GitHub[1] |
Strengths
- Genuinely lightweight — one curl command to a local daemon with a SQLite store; no Kubernetes, no cloud control plane, no account.[2]
- Model-flexible — Claude, Codex, or any OpenAI/Anthropic-compatible endpoint, so it is not locked to one vendor.[2]
- Clean collaboration primitive — @-mention handoffs over an actor-model message queue, with parallel fan-out and isolated workspaces; reviewers called the approach a standout among small frameworks.[5][6]
- MIT licensed with data residency by default — everything stays on the operator's machine.[1]
Cautions
- Development has stalled. No pushes since March 30, 2026, as of June 11, 2026 — roughly ten weeks of silence for a project that shipped 11 releases in its first seven weeks.[1][4]
- Flat adoption. 3,574 stars in June versus 3,507 in April — growth of under 2% in two months, while category leaders grew by tens of thousands.[1]
- Thin community reception. The original Hacker News launch drew 1 point and 0 comments; no substantive critical discussion of the project exists on HN or Reddit as of this writing, which itself signals limited traction.[3]
- No governance or cost controls. No per-agent budgets, approval gates, or audit trail — agents spend whatever the connected API keys allow.[2]
- Branding outruns the product. "One Person Company" suggests organizational structure the product does not have; the legacy repo's own description — "a team of personal agents that collaborate with each other" — is more accurate.[7]
Pricing & Licensing
| Tier | Price | Includes |
|---|---|---|
| Open source | Free | Full product, self-hosted[1] |
Licensing model: MIT — permissive, no paid tier, no managed cloud.[1]
Hidden costs: model API fees for connected providers, plus an always-on machine for the 24/7 daemon.[2]
Competitive Positioning
TinyAGI was listed in the agent-company-platforms comparison, but it competes most directly with personal-agent stacks. Against Paperclip — the category leader, which models agents as employees with org charts, budgets, heartbeats, and board-style governance — TinyAGI offers none of the organizational layer: no reporting lines, no budgets, no approval gates, no audit log.[2]
Direct Competitors
| Competitor | Differentiation |
|---|---|
| Paperclip | Full org-chart governance (roles, budgets, approvals, audit trail) vs. TinyAGI's flat helper team; ~20x the stars and an active release cadence |
| OpenClaw-style personal agents | Single 24/7 assistant vs. TinyAGI's small collaborating crew — TinyAGI's actual peer group[7] |
| CrewAI / multi-agent frameworks | Code-first libraries for developers vs. TinyAGI's installed, UI-driven product[6] |
When to Choose TinyAGI Over Alternatives
- You want 3-5 collaborating helper agents (research, drafts, code, review) on your own machine with minimal setup — and org charts would be overhead.[2]
- You want MIT-licensed local-first software with swappable model endpoints.[2]
- You accept the risk of a project that has not shipped since late March 2026.[1]
Ideal Customer Profile
Best fit:
- Solo operators and tinkerers who want a small personal agent team handling research, drafting, and coding tasks locally
- Developers comfortable forking and self-maintaining an MIT codebase if upstream stays quiet
Poor fit:
- Anyone needing governance, budgets, audit trails, or organizational hierarchy — the defining features of agent company platforms
- Teams or businesses that need an actively maintained dependency or any form of vendor support
Viability Assessment
| Dimension | Assessment |
|---|---|
| Financial Health | No company, no funding, no revenue model — a solo open-source project[1] |
| Market Position | Niche; 3,574 stars and near-zero discussion footprint[1][3] |
| Innovation Pace | Was rapid (11 releases Feb-Mar 2026), now stalled since March 30[4][1] |
| Community/Ecosystem | Small — 505 forks, 62 open issues, 20 watchers[1] |
| Long-term Outlook | Uncertain; depends entirely on whether the solo creator resumes work |
TinyAGI burned bright and brief: an energetic seven-week sprint from Claude Code wrapper to rebranded "one-person company" orchestrator, followed by ten weeks of silence. The codebase remains usable and the MIT license means it can be forked, but as of June 2026 there is no evidence of an ecosystem forming around it.
Bottom Line
TinyAGI is a well-designed, genuinely lightweight personal agent team — and a mislabeled company platform. It has none of the role hierarchy, delegation governance, or cost controls that define the agent-company-platforms category, and its own origin story ("a team of personal agents that collaborate with each other") places it squarely among personal agent stacks.[7] The more pressing issue is momentum: no commits since March 30, 2026, and effectively flat stars since April.[1]
Recommended for: solo tinkerers who want a local, MIT-licensed crew of helper agents today and are prepared to maintain it themselves.
Not recommended for: anyone evaluating agent company platforms — choose Paperclip for the org-chart model — or anyone who needs an actively maintained project.
Outlook: dormant until proven otherwise. If the creator resumes shipping, TinyAGI is a credible entrant in personal agents; if not, it is a historical footnote of the early-2026 agent-team wave.
Research by Ry Walker Research • methodology
Sources
- [1] GitHub API: TinyAGI/tinyagi (accessed June 11, 2026)
- [2] TinyAGI/tinyagi GitHub Repository (README)
- [3] Hacker News: "Tinyclaw: Tiny wrapper of Claude Code that acts as your 24/7 personal assistant" (Feb 10, 2026 — 1 point, 0 comments)
- [4] TinyAGI Releases (latest: v0.0.20, March 26, 2026)
- [5] Jian Liao (@jianxliao) on X — TinyClaw v0.0.4 team-of-agents announcement
- [6] TinyClaw: The Revolutionary Multi-Agent Framework Developers Need — BrightCoding (April 28, 2026)
- [7] TinyAGI/tinyclaw — legacy repository: "a team of personal agents"