An agent-native AI engineering lab.
Tagline: Prompts → Proof
Most engineering workflows and tooling were designed for humans. When the executor becomes an AI agent, we need new primitives:
- Machine-readable work contracts (inputs/outputs, constraints, acceptance gates)
- Replayable evidence (what happened, why it’s safe, how to verify)
- Tool interoperability (glue around existing models/runtimes, not “new models”)
ToolMesh is our umbrella for experiments, open-source components, and productized delivery that make agents reliable at enterprise quality.
ReceiptCI / Proofpack — a proof-carrying CI gate for agent work.
Every agent change ships with a work contract + receipts (evidence) + a replay path CI can verify deterministically.
We ship small “test rigs” through the same agent-first flow, then productize the reusable parts:
flowchart LR
A["Work Contract (spec)"] --> B["Agent Execution (tools)"]
B --> C["Receipts (evidence)"]
C --> D["CI Replay/Verify"]
D --> E["Ship + Observe"]
- Templates & pipelines for agent-native delivery
- Reference projects (our “showroom”)
- Curated resources: awesome-ai-agents
- Hub repo + pitch notes: toolmesh
Agent-to-Prod Sprint (2–4 weeks): we help teams install an agent-first SDLC (contracts, receipts, quality gates) and deliver the first production use-case.
Open issues/PRs in any repo, or start with the showroom hub (coming soon).