What we do and why we built it differently.
Most agent harnesses compete on scale - more skills, more agents, more stars. Agent Foundry competes on quality and transparency.
Every skill in our catalog is original MIT-licensed work, verified by three automated quality gates that run on every commit. We have a live knowledge graph of the codebase itself - not a static list, but 766 connected nodes extracted via tree-sitter AST. Every one of the 14 harness adapters passes the installer + indexer smoke test, and the same skills ship through npm, PyPI, and the official MCP registry.
Every skill passes static eval, frontmatter validation, and an external-reference scan. No adapted, copied, or unverified content.
The knowledge graph shows every code symbol, its connections, and its community cluster. Updated on every commit.
Smoke tests are in the repo. Quality gates are in the repo. Graph pipeline is in the repo. Everything is auditable.
MIT license, local daemon, no telemetry, no cloud. You supply the API key. We never see your prompts.
We don't ship hundreds of unverified skills adapted from third-party sources. We don't run a SaaS subscription. We don't claim "self-improving" when the improvement system produces mostly noise. We don't inflate metrics.
We ship honest software with measurable quality guarantees.