Open standards · MIT
Open standards for runtime governance
Rends authors and ships the protocols that runtime governance needs to exist as an interoperable layer — not a single vendor's product. Implement them, fork them, or hold us accountable to them.
Protocols
What we publish
Rends Policy Language
RPL is the open policy language that Rends Govern evaluates. It compiles to deterministic decisions, content-addresses every revision with SHA-512, and is small enough that a third party can implement a validator and evaluator from the spec alone — no Rends dependency required.
- Spec & conformance suite:
- github.com/eishops23/rends-policy-language
- Current version:
- v0.x — pre-1.0, spec stabilizing
- License:
- MIT
- Reference implementation:
- Agent SDK
Agent Audit Log Schema
AALS is an open record format for agent governance decisions: canonical JSON serialization, SHA-512 record hashes and hash-chain linkage, so anyone holding a chain of AALS records can verify it with the spec and a hash function, without calling whoever produced it.
v1.0 (published) records each decision and its violations by policy and rule name or ID. It does not pin policy revisions. v1.1 (draft for public comment) adds revision pinning: a hash of the whole rule set in force and, for each rule that applied, its exact revision and content hash. It also records the decision's effect, including escalation, and how an escalation ended.
Rends does not yet export AALS records. Its own decision records use their own format, described under Policy anchoring.
- Schema & fixtures:
- github.com/eishops23/agent-audit-log-schema
- Regulatory mapping:
- EU AI Act Art. 12, NIST AI RMF MANAGE-2.3, ISO/IEC 42001 §A.6
- License:
- MIT
- Reference implementation:
- Agent Governance Stack whitepaper
Policy anchoring
Every Rends decision record pins the policy in force when the decision was made: a SHA-512 hash of the whole rule set in scope, including the rules that allowed, and, for each rule that applied, its exact revision and content hash. Rule revisions are immutable and content-addressed, so a rule edited afterwards no longer matches the hash its decisions recorded.
- Where it ships:
- The Rends decision chain (hash-chained decision records)
- Where it shows up:
- Every decision record since 28 September 2026. Not the older audit trail: its rows do not carry the pin.
Federation-ready by design
Rends is designed so independent deployments can eventually exchange verifiable decision records. Some of the building blocks already ship: canonical JSON serialization, per-org SHA-512 hash chains, and content-addressed rule revisions, which every Rends decision record has pinned since 28 September 2026. Standalone verification does not ship yet. Rends does not export decision records with their hashes, so a third party cannot yet check a decision chain without Rends. The public verifier at /verify/chain checks a demonstration audit-trail chain in your browser.
A formal exchange protocol — RFC-0 — is planned for a draft in 2026. We'll publish the draft here when it's ready and invite implementations before we ship our own. If you want to be in the conversation early, the Builders invitation below is the right door.
Get involved
Three ways in
For operators
Be design partner #1 for runtime governance. We're picking customers, not the other way around.
Apply →For builders
Implement the specs. Open an issue if the spec is unclear. We'll co-develop the conformance suite with you.
GitHub →For standards reviewers
Review the schema, the canonicalization rules, the hash-chain construction. Issues, PRs, and prose critiques all welcome.
Reviewers' issues →Already implementing one of the specs? File an issue at RPL issues or AALS issues — we'll track conformance and link your implementation here.
Spec repos currently live under our personal GitHub. We plan to move them to a github.com/rends-ai org as soon as the org name is available — the current URLs will redirect.
Read the long-form treatment in The Agent Governance Stack and our essays at /research.