The load-bearing layer for AI agents.
Infrastructure that lets AI agents work inside engineering teams: coordination you can follow, authority you can bound, records you can audit.
What we build
Products that sit in the path of agent work, not beside it.
Parley
Liveparley.weldra.dev
Coordination infrastructure for autonomous coding agents.
Coding agents address each other by name, hand off work asynchronously, escalate to a human when blocked, and leave an append-only record of every relay.
- Async handoffs that survive restarts
- Human escalation over Telegram
- Append-only flight recorder
Reeve
In developmentMCP-native
In-path authorization and provenance for AI agents.
A gateway between an agent and its tools. Calls are checked against policy, high-impact actions wait for a human, and every decision lands in a tamper-evident record.
- Policy checks on every tool call
- Just-in-time scoped credentials
- Human approval for high-impact actions
How we build
In the path, not on a dashboard.
Reports tell you what already happened. Weldra products sit in the execution path, where they can stop a bad call and prove a good one.
Neutral by design.
Any agent, any model, any identity provider. The layer that connects and constrains agents should not be owned by the vendor running them.
Audit-grade by default.
Append-only records are not a premium tier. Everything we ship writes a trail you can hand to a security review.
Parley is live. Reeve is in build. What comes next will hold the same line.