Memnox sits between your AI agents and your systems and returns one of three answers, allow, block, or ask a human, before anything runs. It understands code; it never writes it.
Verdicts come from a deterministic policy engine. Same action, same decision, every time. Security needs guarantees, not probabilities.
TypeScript, Python, Go, Java, Rust and Swift clients all ask the same runtime, so a rule holds wherever it is called from.
No account, no API key, no network call. The solo path never talks to a server that isn't on your laptop. Postgres and Redis when one process stops being enough.
The first run decides nothing. Read the audit log, then re-run with --enforce once the calls look right.
Every decision appends to a hash-chained log carrying the policy version that decided it. memnox audit verify shows whether a single record was touched.
Read every line of the thing governing your agents. Runs on a laptop, in your cluster, or on a network that never leaves the building.
npx memnox setup# Wrote starter policies to memnox.policies.yaml# Detected: payments, database migrations, CI/CD# Restart your editor and it is governed.
One command scaffolds policies from what it detects in the repo, registers a local agent and starts the runtime.
The gate that decides what an agent may do is Apache-2.0 and runs on your machines. The understanding of your organization is what the cloud keeps current. Each is complete on its own, and adopting one never holds the other hostage.
Everything that decides is open: the same gate whether one laptop runs it or the whole company does.
Everything that understands is managed: the picture of the business that every person and every agent works from.
Agents and the gate can run entirely in your environment while the cloud manages policies, the fleet and the picture everyone shares. Nothing about the local path changes when the organization adopts the cloud, and the record is always yours to export.