Teamplane gives every teammate and every agent a signed mandate, a lease on what it touches, a permissioned memory it may read, and an auditable merge gate.

Every decision is reproducible from the event log.
Ed25519 capabilities bound to one principal and one task. Derivation only narrows. Verification works offline; revocation is an event.
Deny > ask > allow, default deny, plus typed leases with fencing tokens. No model is consulted at decision time.
Typed events and curated memory fragments with provenance and ACLs. The team sees conclusions, not conversations.
# build and start a seeded playground npm install && npm run build && npm link teamplane playground --open # or run the executable proof npm run demo # verify the audit chain teamplane audit verify
ask until an owner approves, once
DeepSeek Harness, Claude Code, Cursor and Codex connect over MCP — stdio or Streamable HTTP.
The ten tp_* tools cover leases, memory, approvals, events and submit.
“The model keeps the interesting job — planning, coding, judgement — and loses the two jobs it is worst at: being an authority and being a lock.” docs/design-notes.md
# DSH profile patch - id: mcp-teamplane name: '@deepseek-ai/dsh-mcp-client' config: serverName: teamplane transport: stdio command: teamplane args: ['mcp'] env: TEAMPLANE_URL: http://127.0.0.1:7317 TEAMPLANE_TOKEN: !!js process.env.TEAMPLANE_TOKEN
The design notes cite the prior art it builds on — Collaborative Memory, MERGEGYM, ScopeGate, DelegateGuard and DeepSeek Harness's Super-agent — and state what would falsify the design.