Skip to main content

Covenant: open agent-native operating layer

Covenant is the open, local-first coordination layer for agentic software. It gives humans and agents eight host-level primitives (intent, runtime, memory, identity, permissions, communication, a compositor, and on-chain settlement) so they can safely share one computer. Try the interactive sandbox, read the documentation on the eight primitives, review the development roadmap, or read the technical whitepaper.

An operating system that builds itself

Not a chatbot. Not an agent framework. An open operating layer where every agent runs under a signed grant, every action leaves a receipt, and the system itself is built in the open by the autonomous infrastructure it provides.

eight host-level primitives

One computer, safely shared by humans and agents

what you plug in

Two ways agents meet the real world safely

Guard

trust layer

Before your agent pays or trusts another agent, it checks an on-chain track record, confirms a real identity, and verifies a signed claim. A zero-install MCP for Claude and Codex.

Explore Guard →

Trading

governed execution

Brokerages are opening to agents with no real limits. Covenant is the policy gate in front of the order: caps enforced before it's placed, and every decision anchored onchain where anyone can check it.

See governed trading →

built in the open

The infrastructure builds itself

Covenant is written, in the open, by the autonomous loop it ships. Every commit runs under a signed grant, is gated before it executes, and leaves a receipt in a hash-chained audit. The system holds itself to the rules it sells you.

Watch the live build log →

ecosystem

Speaks the protocols agents already use

Settlement on Solana, pay-per-call over x402, tools and agent-to-agent traffic over MCP and A2A, and a growing set of partner products that reach agents as capability-scoped, attested tools.

See the partners →