"From Prompt to Protocol"
We are proud to announce the first stable release of ReasonKit Core, marking the shift from experimental AI to engineered cognitive pipelines.
ReasonKit is no longer just a Rust binary; it's a configurable engine.
- YAML Protocols: Define how the AI thinks in
protocols/thinktools.yaml. - Custom Profiles: create your own reasoning chains in
protocols/profiles.yaml. - Hot-Reload: Change a protocol description, see it update instantly.
The engine now ships with 5 stabilized cognitive modules:
- GigaThink (
gt): Divergent thinking (10+ perspectives). - LaserLogic (
ll): Deductive reasoning & fallacy detection. - BedRock (
br): First principles decomposition. - ProofGuard (
pg): Triangulated verification. - BrutalHonesty (
bh): Adversarial critique.
Stop building black-box agents.
- New
examples/rk_langchain_agent.pydemonstrates how to use ReasonKit as a verifying backend for Python agents. - Agents can "offload" hard reasoning to ReasonKit and get structured, auditable JSON back.
We are proposing a new community standard for AI integrity in BENCHMARKS.md.
- Moved beyond "vibe checks" to structural integrity metrics.
- 5 standard "Hard Reasoning" prompts to test depth, breadth, and grounding.
- Performance: All core reasoning loops validated at <5ms latency (excluding LLM inference).
- Stability: Resolved JSON parsing issues in the Python integration.
- PowerCombo: Marked as
experimentalwhile we refine the composite logic.
cargo install reasonkit-core
# or build from source
git clone https://github.com/reasonkit/reasonkit-core
cd reasonkit-core
cargo build --release