Our mission is to make agents worthy of enterprise trust. Praetor is alpha: these films distinguish the goal, component-level engineering, and one recorded API experiment.
The mission 0:29
Can enterprises trust agents with important work? This is the long-term problem we are working to solve.
Brand vision · Alpha; a mission still being built
Agents can act. Can enterprises trust them? Trust has to be earned. Action by action. Our mission: make agents worthy of enterprise trust. A long-term mission. Built action by action.
The engineering 1:58
Exact request authority, stale-worker fencing, uncertain outcomes, and signature checks that reject altered evidence.
Mechanism illustrations · Component tests, including in-memory repositories and mocked HTTP
Approval binds to an exact request; changing an argument changes its digest. Reusing an idempotency key with different arguments produces a conflict. Expired leases reject stale commits. A timeout cannot establish a remote outcome: preserve UNKNOWN, query read-only, and reschedule when it remains uncertain. Rehashing altered evidence can satisfy structural checks, but verification against the trusted signing key fails. A signature establishes integrity and signing origin, not remote truth. These are component-level checks, not production-readiness evidence.
The live API experiment 0:38
Real GitHub API, Gateway and PostgreSQL: inject a lost reply, restart the worker, and recover the existing pull request.
Animated reconstruction of a CI run · Injected faults, no LLM, not a production incident
An action is proposed. The proposing agent cannot approve itself. A separate identity approves the exact request. GitHub creates the PR; the test proxy drops the reply and the worker is killed. COMPLETION_UNKNOWN is persisted. After restart, a read-only lookup finds one matching PR by action marker and resumes to COMPLETED, with one create call in the shown scenario. Zero or multiple matches remain unknown. This uses a sandbox repository and injected faults, with no LLM in the loop. It does not establish a universal exactly-once guarantee.
English on-screen text with an instrumental score, no voiceover. English captions and text transcripts are available.
The technical film is supported by 81 selected component checks at source revision 1799a15c. These are recorded results, not a full-suite pass or production validation. The authority and fencing checks use in-memory repositories; outcome checks include mocked HTTP and controlled repositories. The signed bundle is synthetic: it demonstrates integrity checks, not an external action.
Read the quick start or inspect the GitHub pilot boundary. Review the current readiness and limits before evaluating production workloads.