Work
Open-source infrastructure for trustworthy autonomous agents: the trust layer, the protocols, and the runtimes around it. Click any project for the full story.
- Sardis → Working
A policy engine that approves or denies an AI agent's payment against a signed mandate before any money moves, and proves every decision afterward.
- OAPS / AICP → Draft spec, working reference
An open protocol that records who authorized an agent action, what limits applied, and what proof exists, shared across MCP, A2A, and payment systems.
- OSP → Spec complete, SDKs working
An open protocol that lets an agent discover, provision, and pay for a service in one HTTP call, then rotate the credentials it gets back.
- FIDES → Core working, service partial
A control layer that verifies an agent's identity, capabilities, and authority before an action runs, then logs tamper evident proof of the decision.
- Agentbox → v0.1 working
Run an autonomous agent on your own machine with a phone approval gate for risky commands, instead of choosing between full trust and a separate box.
- Capsule → Core working, adapters partial
One TypeScript interface to run agent code across Docker, Cloud Run, Vercel, and Neon, where every adapter declares exactly what it supports.
- Switchboard → Working
A local daemon that gives parallel coding agents one shared, tamper evident ledger, so they stop redoing work and faking "done".
- Vela → Phase 0 working
A control plane that scores and gates whether a code change, human or agent written, can be trusted before it merges.
- Yula → Working core, hardening
A control layer that classifies every agent tool call by how reversible it is, signs the authority behind it, then blocks it, holds it for approval, or runs it, with a timeline to verify, undo, or replay.
- agit → Core working
Git for AI agent state. Commit, branch, diff, and bisect an agent's memory and actions to replay and debug how it went wrong.
- better → v0.1 working
A Node.js package manager with a Rust core that installs far faster by using macOS copy on write and deduplicating files across projects.
- Maestro → Core working
A GitHub app that scores a pull request 0 to 100 across behavior, risk, performance, UX, and security, and returns ship, review, or wait.
- Sardis Company Builder → Working, deployed
An AI agent that researches, validates, and specs a company by paying real services per call over MPP micropayments, planned live by Claude.
- Synthesis → Working demo
A runnable demo of two agents completing a payment with policy checks, trust scoring, and an on chain anchored audit trail, all verifiable over HTTP.