Make_Skills
A multi-agent platform for building, raising, and running personal AI agents. Open-source first, hosted second.
A multi-agent platform for building, raising, and running personal AI agents. Each user has a clan of specialized agents that learn from how they work.
What it does
- Memory persists across sessions and across agents (LanceDB-backed semantic memory).
- Skills accumulate as you work — markdown files agents read, sometimes promoted to tools.
- Subagents specialize: orchestrator, planner, researcher, plus custom specialists. Each has its own role and contract.
- Observability covers agent communication tracing, decision audit trails, and Grafana dashboards.
- Self-host on your machine, or use a hosted instance.
Pillars
Pillar 0 — Foundational
Open-source-ready and dual-mode (self-host AND hosted-multitenant)
from day one. Tenant abstraction, auth interface, repo strategy.
Pillar 1 — Build agents
Create a clan of 3D creatures. Each subagent has its own board
and evolves as it gains skills.
Pillar 2 — Make skills together
Collaborative human-agent skill-building. The recorder extracts
patterns; the curator proposes new skills; users approve;
agents grow sharper.
Pillar 3 — Observability
3a. Agent comms (multi-agent flow tracing, cost per skill)
3b. Grafana dashboards
3c. Knowledge library (extract → review → canonical graph,
modeled on PROVES_LIBRARY)Where to go next
Status
Self-host runs end-to-end today. Hosted-multitenant infrastructure (auth, tenant abstraction) is in progress. Roadmap at /roadmap on humancensys.com.
License: Apache 2.0.