← Back to research
·9 min read·industry

ClawCompany

ClawCompany is an open-source "AI company OS" that ships 38 predefined roles across 6 company templates with tiered model routing it claims is 30x cheaper than all-Opus. MIT licensed, 574 GitHub stars as of June 2026 — down from earlier counts — with no commits since April 25, 2026.

Key takeaways

  • 574 GitHub stars as of June 11, 2026 — a net decline from the 749 recorded by OSS Insight in early April and the 899 this site's April hub recorded, an unusual downward trajectory consistent with a star purge or recount
  • Tiered model routing is the signature idea: professional work gets Opus/Sonnet, routine work gets Flash-Lite, with a project-claimed 30x cost reduction versus running everything on Opus
  • No commits since April 25, 2026 (~47 days as of this writing) and only two visible commit authors — the project's biggest open question is whether it is still maintained

FAQ

What is ClawCompany?

ClawCompany is an open-source, locally run "AI company OS" that organizes AI agents into predefined company roles (38 roles, 6 templates), positions the human as Chairman, and routes tasks to different LLMs by complexity. MIT licensed, installed via npx clawcompany.

How much does ClawCompany cost?

The software is free (MIT, self-hosted). You pay model API costs; the project claims roughly $0.06–$0.12 per completed mission with its tiered routing, versus several dollars on premium models alone — project figures, not independently verified.

What models does ClawCompany use?

It routes by task complexity — Claude Opus/Sonnet for professional work like CEO-level decomposition, Gemini Flash-Lite for routine work — and an April 2026 commit added GPT-5.5 as an Opus fallback.

How is ClawCompany different from Paperclip?

Paperclip gives you raw org-chart orchestration primitives and lets you build your own company; ClawCompany ships pre-built role libraries and company templates with cost-tiered model routing out of the box — but with a fraction of the adoption and, as of June 2026, stalled development activity.

Executive Summary

ClawCompany is an open-source "AI company OS": you act as Chairman, and an autonomous AI team — drawn from 38 predefined roles across 6 company templates — decomposes missions into work streams, executes them, and delivers finished output, running locally via npx clawcompany.[1] Its signature idea is tiered model routing: professional work (strategy, decomposition, complex analysis) goes to Claude Opus/Sonnet, routine work goes to Gemini Flash-Lite, which the project claims is "30× cheaper than all-Opus."[1] A 4-layer memory system (session context, compressed archives, company knowledge, Chairman preferences) injects roughly 400 tokens of context per mission so the company "remembers everything — locally."[2]

Two things demand a clear-eyed note. First, development has stalled: the last push to the repository was April 25, 2026 — about 47 days of silence as of this writing — with only two visible commit authors across recent history.[3][4] Second, the star count has gone backwards: the GitHub API reports 574 stars as of June 11, 2026, while OSS Insight's early-April snapshot recorded 749 and this site's April 2026 category hub recorded 899.[3][5] A net decline of this size usually indicates a GitHub star purge (removal of bot/fake stars) or a recount; the cause cannot be determined from public data, but the discrepancy is real and worth flagging in a category OSS Insight already criticized for star counts that "reflect marketing hype rather than real deployment."[5]

AttributeDetail
CreatorClaw-Company GitHub organization; no disclosed team or company
CreatedMarch 14, 2026[3]
FundingNot publicly disclosed
GitHub Stars574 (as of June 11, 2026; down from earlier counts)[3]
LicenseMIT[3]
Last pushApril 25, 2026[3]

Product Overview

ClawCompany installs as a local TypeScript/Node.js application. You pick a company template, optionally customize roles, then issue missions as Chairman. A leader agent (CEO) recalls past missions and your preferences from memory, decomposes the work into streams — researchers gather data, analysts build models, writers format reports — and returns a completed deliverable.[1] Recent releases added document export to PPT/Word/PDF (v0.60.0) and a subagent tool (v0.53.0).[4]

It is independent software, not built on OpenClaw — the README instead positions OpenClaw as an alternative requiring more configuration, claiming an example mission completes for $0.06 versus "$0.40+" on OpenClaw.[1]

Key Capabilities

CapabilityDetail
Company templates6: Default/General (9 roles), YC Startup (7), Trading Desk (7), Research Lab (5), Software Dev (6), Harness Builder (3) — 38 roles total[1]
Tiered model routingOpus/Sonnet for professional work, Flash-Lite for routine; project-claimed 30x cost reduction vs all-Opus[1]
4-layer memoryLayer 1 session context, Layer 2 compressed archives, Layer 3 auto-categorized company knowledge, Layer 4 Chairman preferences; ~400 tokens injected per mission[2]
Agent tools9 built-in: Web Search, Web Fetch, Price Feed, Browser Use, Shell, Filesystem, HTTP, Code Interpreter, Memory Search[1]
Human-in-loopChairman model — the human initiates missions and reviews outputs, retaining executive authority[1]

Technical Architecture

Fully local and self-hosted — there is no cloud service or account. Requires Node.js 20+:[1]

npx clawcompany

Model access is multi-provider: Claude Opus/Sonnet for high-tier work, Gemini Flash-Lite for routine tasks, and — in the final commit before the quiet period — GPT-5.5 wired in as an Opus 4.7 fallback ("providers: add GPT-5.5 as Opus 4.7 fallback," April 25, 2026).[4]

Key Technical Details

AspectDetail
DeploymentLocal CLI/app via npx clawcompany, Node.js 20+[1]
LanguageTypeScript[3]
ModelsClaude Opus/Sonnet, Gemini Flash-Lite, GPT-5.5 fallback — routed by task tier[1][4]
Latest releasev0.61.0 (April 17, 2026 — Opus 4.7 with auto-migration)[4]
Open SourceYes — MIT[3]

Strengths

  • Cost optimization as architecture — Tiered model routing was the first serious attempt in this category to treat model cost as a first-class design concern; the principle (match model capability to task complexity) holds even as the specific 30x figure is a project claim[1]
  • Fastest time-to-deploy in the category — 38 predefined roles across 6 templates mean you pick a structure and go, rather than designing an org chart from scratch[1]
  • Local-first memory — The 4-layer memory system keeps company knowledge and Chairman preferences on your machine, with a lean ~400-token injection per mission rather than stuffing context windows[2]
  • Clean issue tracker — 1 open issue as of June 11, 2026, suggesting the small user base isn't hitting (or reporting) breakage[3]
  • MIT licensed — Permissive license, fully self-hosted, no account or vendor dependency[3]

Cautions

  • Development has stalled — No pushes since April 25, 2026, roughly 47 days as of this writing. For a project that was six weeks old when it went quiet, this is the dominant risk factor[3]
  • Star count declined — 574 stars in June versus 749 in OSS Insight's April snapshot (and 899 in this site's April hub) — consistent with a star purge or recount, and a caution sign in a category already flagged for inflated star counts[3][5]
  • Effectively a single-maintainer project — Recent commit history shows only two authors: the ClawCompany organization account and an account named "claude," with no disclosed team, company, or funding behind either[4]
  • Unverified cost claims, inconsistently stated — The README cites $0.06 per mission; the website cites ~$0.12 per mission (5 roles, 4 models) versus $3.50 all-premium. Both support the "30x" framing, but neither is independently benchmarked and they don't match each other[1][2]
  • Pre-1.0 software — Latest release is v0.61.0; export, memory compression, and subagent features all landed within weeks of the activity stopping[4]
  • Minimal community footprint — OSS Insight gave it minimal discussion next to Paperclip, and no substantive third-party reviews, HN threads, or critical coverage were found as of June 2026[5]

Pricing & Licensing

TierPriceIncludes
Open sourceFreeFull platform, all 38 roles and 6 templates, self-hosted[1]

Licensing model: MIT — permissive open source, no paid tier, no managed cloud, no commercial entity to buy support from.[3]

Hidden costs: Model API spend (Anthropic, Google, OpenAI keys) is the real cost; the project's own per-mission estimates range from $0.06 (README) to ~$0.12 (website), unverified.[1][2]


Competitive Positioning

Direct Competitors

CompetitorDifferentiation
PaperclipThe category leader (69k+ stars, active biweekly releases) — raw org-chart orchestration primitives, agent-agnostic, build-your-own-company. ClawCompany trades that flexibility for pre-built templates and cost routing, but Paperclip kept shipping while ClawCompany went quiet
Oh-My-ClaudeCodeWraps Claude Code with multi-agent team coordination; locked to one agent ecosystem where ClawCompany is multi-model
EdictTang Dynasty-inspired governance with an institutional AI veto layer; strongest safety model in the category, built on OpenClaw
MindStudioCommercial no-code visual builder for non-technical operators; SaaS where ClawCompany is local and free

When to Choose ClawCompany Over Alternatives

  • You want a working multi-role AI company in one command, not an orchestration framework to configure
  • Model API cost is your binding constraint and tiered routing matters more to you than ecosystem maturity
  • You require fully local operation — memory, preferences, and company knowledge never leave your machine
  • You are comfortable adopting (or forking) a project that may no longer be maintained

Ideal Customer Profile

Best fit:

  • Solo operators and tinkerers who want pre-built role templates (research, trading-desk analysis, software dev) with minimal setup
  • Cost-sensitive experimenters running many missions on a budget
  • Developers studying tiered model routing or 4-layer memory design — the codebase is a compact MIT-licensed reference implementation

Poor fit:

  • Anyone who needs an actively maintained dependency — 47 days of silence makes this disqualifying for production use
  • Teams wanting vendor support, SLAs, or an identifiable company behind the software
  • Users who need deep org-chart customization, multi-company portfolios, or a broad agent-adapter ecosystem (choose Paperclip)

Viability Assessment

DimensionAssessment
Financial HealthNo disclosed funding, team, or revenue model; free OSS with no commercial entity
Market PositionFourth-tier in its category — 574 stars versus Paperclip's ~70k; minimal third-party coverage[3][5]
Innovation PaceWas rapid (v0.36.0 → v0.61.0 in roughly six weeks), then stopped entirely on April 25, 2026[4][3]
Community/Ecosystem72 forks, 62 watchers, 2 visible commit authors, 1 open issue — small and quiet[3][4]
Long-term OutlookPoor unless activity resumes; the ideas (tiered routing, template libraries) are likely to outlive the project itself

ClawCompany shipped genuinely good ideas fast — OSS Insight measured it at 39 stars/day in its first 19 days — and then went silent six weeks in.[5] The repository is not archived, so a return is possible, but every viability signal (commit activity, star trajectory, maintainer count) currently points the wrong direction.[3]


Bottom Line

ClawCompany is the category's best argument that cost routing belongs in the architecture — and its cautionary tale about single-maintainer velocity. The template-maximalist approach (38 roles, 6 companies, one command) delivered the fastest out-of-box experience in the agent-company space, but the project has not seen a commit since April 25, 2026, and its star count has declined rather than grown.

Recommended for: experimenters who want a free, local, template-driven AI company and accept abandonment risk; developers mining it for tiered-routing and memory-design patterns.

Not recommended for: anyone building on it for real work — the maintenance signal is the worst in the category.

Outlook: quiet, possibly dead. The tiered model routing idea is already being validated elsewhere in the category; whether ClawCompany itself revives is a question only its anonymous maintainer can answer. Re-check the commit log before adopting.


Research by Ry Walker Research • methodology