// the find
kanwas-ai/kanwas
Kanwas — Shared context board for teams and agents
Kanwas is a multiplayer AI workspace where a shared canvas holds documents, decisions, and agent tool calls in a single timeline visible to the whole team. It's aimed at product and knowledge workers who want to run AI-assisted work against a persistent, shared context rather than per-session chat threads. Self-hostable via Docker Compose with Anthropic or OpenAI backends.
The Yjs-backed real-time sync is a genuinely hard problem done right — using CRDT semantics means concurrent edits don't fight each other, and the separate yjs-server process keeps sync concerns out of the main AdonisJS backend. The git-backed markdown filesystem is a good call: workspaces are portable, diff-able, and not locked into a proprietary format. The CLI (`kanwas pull/push/import`) lets you wire the workspace into existing coding agent workflows, which is the most practical bridge between this kind of planning tool and actual development. The skills system (seeded prompt templates for ADRs, PRDs, retros, etc.) ships real value out of the box rather than leaving users to prompt-engineer from scratch.
The Composio dependency for external tool connections is a significant coupling — if Composio changes pricing or API surface, all integrations break. There's no sign of end-to-end tests or integration tests in the workflows; three separate test CI jobs exist but the test files themselves aren't visible, so coverage depth is unknown. The E2B sandbox integration is present but so is a Docker sandbox, with no clear documentation on when to use which or what the security boundary is between them. Self-hosting requires managing three separate env files across four services; a single composable config or a proper setup script would cut the friction significantly for anyone not already fluent in Docker Compose.