Tools for agents who stay.
Most agent tooling assumes a task: spin up, execute, vanish. This org builds for the other case — agents that persist. Memory that survives the session. A voice that holds relationships across months. Policy and permissions that travel with the agent. The discipline of keeping long-lived agents honest.
| pali | Semantic memory for AI agents — git-backed, HNSW-indexed, scoped namespaces. Formerly memory-mcp. |
| dione | Discord presence for AI agents — channel routing, access control, delivery contracts, and live personas |
| prodagent | Agent authorization — shell parsing, command knowledge, and policy hooks for safe tool execution |
| agent-sandbox | OCI image for running AI coding agents in microsandbox — Claude Code, Codex, Gemini, OpenCode |
| interlockutor | Claim-once work queue and broadcast event bus with interlocking safety guarantees |
| lacuna-skills | Harness-neutral canonical skills for persistent agents |
| claude-marketplace | Skills, hooks, and MCP server plugins for Claude Code |
| construct-nameplates | Opt-in, self-authored nameplates for constructs |
| entmoot | Conversation-tree context service — tends the trees so constructs can see the forest |
mcp-session — bounded session management for MCP servers · celastrina — custom Bazzite image with nvidia desktop and Lenovo Yoga 9 laptop builds · gossamer — system observability · catena — a MUD engine in Rust