Pinchy

Pinchy is a self-hosted AI agent platform for teams. It centers on one agent per role, with boundaries that control who can see what, which data each agent can access, and which tools it can use.

The product describes those boundaries as the core of the system. It includes role-based access for people, per-agent permissions, group-based visibility, default-deny access rules, scoped plugins instead of raw shell access, and a signed audit trail that is described as tamper-evident, append-only, and exportable to a SIEM. The page also says agents can be given their own knowledge base, with cited sources in answers, and that usage analytics are available by source, agent, user, and timezone.

Pinchy is intended for team use cases across business software and external tools. Examples shown on the page include a quote-drafting agent for sales, a support writer that drafts replies in Gmail or Outlook, an onboarding guide for people operations, and an invoice bot for finance. It is described as model-agnostic and as working with OpenClaw, Odoo, Gmail, Outlook, Telegram, Ollama, OpenAI, and Anthropic. The deployment section says it runs self-hosted, with one Docker Compose command on any Linux box, and that pre-built GHCR images are available. The page also says it can run fully air-gapped and, when paired with local models via Ollama, completely offline.

Licensing and pricing are stated directly on the page. Pinchy’s core is licensed under AGPL-3.0, the codebase is open source, and the free Community edition is fully functional for small teams. A paid Pro tier adds team-governance features such as groups, agent access control, and usage analytics via a license key. The page also mentions a 30-day trial.