Most agent frameworks let the model both do the work and decide what happens next. That is fine for demos and brittle in production. Kraken splits the job: models do the work, deterministic rules decide what happens next. Every decision is inspectable, versioned, replayable, and free of stochastic drift.
Fathom · Nautilus · Stargraph · Discord · Discussions
flowchart TD
SG["<b>Stargraph</b><br/>stateful agent-graph framework<br/>rules route transitions, not the LLM"]
NA["<b>Nautilus</b><br/>policy-first data broker<br/>plan · route · enforce · attest · audit"]
FA["<b>Fathom</b><br/>deterministic reasoning runtime (CLIPS)<br/>YAML rules · microsecond eval · zero hallucinations"]
SG --> FA
NA --> FA
Fathom is the decision engine. Nautilus and Stargraph are two applications of it. Use Fathom alone to add a deterministic decision layer to any agent; use Nautilus to broker data access under policy; use Stargraph to build whole auditable agent graphs.
🧭 Fathom — the engine
Deterministic reasoning runtime for AI agents. Define rules in YAML, evaluate in microseconds, zero hallucinations. Built on CLIPS.
uv add fathom-rulesReplaces the LLM-as-router pattern for any decision that can't be a maybe — policy enforcement, data routing, guardrails, eligibility.
🌊 Nautilus — the data broker
Policy-first data broker for AI agents. One call plans, routes, enforces, attests, and audits — so agents never touch data directly.
uv add nautilus-rkmBuilt on Fathom. For regulated data (HIPAA / NIST / air-gapped) where every access must be provable and audited.
⭐ Stargraph — the agent framework
Stateful agent-graph framework with deterministic governance. Compose LLMs, ML models, tools, and rules into auditable, replayable graphs.
uv add stargraphBuilt on Fathom. For environments where auditability, determinism, and provenance matter more than ecosystem size.
- Deterministic by construction — the decision layer is rules, not a sampled token stream.
- Auditable & replayable — every routing decision is inspectable and reproducible.
- Provenance-typed — decisions are made over typed, attributable facts.
- Built for regulated & cleared workloads — DoD, healthcare, finance, air-gapped.
We want these projects to be easy to find and even easier to contribute to. Every repo
has runnable contributor docs, structured issue forms, good first issue on-ramps, and CI
that auto-fixes formatting instead of failing your build.
- Start at any repo's
CONTRIBUTING.md - Browse
good first issueacross the org - Chat with us on Discord or ask in Discussions
- Read our Code of Conduct and AI contribution policy
