Back to Gap Map Product / UX / Agent identity & authorization

OpenFirma

Firma AI
open source / Overall score: 2.2

OpenFirma puts a local sidecar between AI agents and the outside world and checks every outbound call against Cedar policies before it leaves. An authority issues short-lived capability tokens to each agent, the sidecar injects credentials only into an allowed call, and every decision is written as a signed audit event. Firma AI builds it in Rust.

Openness

5 medium confidence
5.0
license
GPL-3.0 (root LICENSE; the Cargo workspace declares GPL-3.0-only)+Apache-2.0 (firma-protobuf crate)
source
public
core features withheld
no — the 1,235-path tree at 1d79361 has no ee/, enterprise/, commercial/ or proprietary/ path and holds the sidecar, the Mini Authority, the sandbox runner and the CLI; the hosted Firma Team control plane at app.firma.ai is a separate service the CLI can log in to, and the README says the Mini Authority or an operator's own implementation of the authority interface can serve
commercial
Firma Team control plane (a hosted SaaS with workspaces and an agent registry, also offered for on-premise deployment)

OpenFirma's sidecar, authority, sandbox runner and command-line tool are published under GPL-3.0, and the whole enforcement path runs locally without an account. Firma AI's hosted Team control plane, which adds shared workspaces and an agent registry, is a separate service; the published authority can take its place, or an operator can write their own.

Adoption

1 low confidence
1.0

OpenFirma ships through its own install script and Homebrew tap, neither of which publishes install counts, and its Rust crates are not published to crates.io, so GitHub stars are the only signal. A star is not a use, so this shows attention rather than deployments.

Capability

4 medium confidence
4.0

OpenFirma gives each agent a signed capability token, checks every outbound call against policy before it runs, lets a person approve local execution, and signs every decision it records. Its authority grants each agent a policy-narrowed set of actions, but there is no chain for one agent to hand narrower authority to another.

Verified 2026-11-20, except some scores, still awaiting re-confirmation