The AI-native causal decision engine
Knowledge graphs, deterministic execution, and DAG-based receipts. Build AI systems that persist and explain every decision they make.
What changes with Cruxible
LLM agents are powerful but unreliable decision makers. Cruxible grounds them with structured knowledge, reproducible execution, and provenance for every result.
LLM agents alone
- Relationships shift depending on how you ask
- No structured memory between sessions
- Results vary between identical prompts
- "Trust me" — no audit trail
- Constraints checked by vibes
- Learns nothing from outcomes
With Cruxible
- Explicit knowledge graph you can inspect
- Persistent entity store across runs
- Deterministic execution, same input → same output
- DAG-based receipt for every decision
- Declared constraints validated before results
- Feedback loop calibrates edge weights over time
Four primitives, one loop
Every Cruxible Core deployment follows the same pattern: configure your domain, ingest data, query for decisions, and feed outcomes back.
Example entity graph — products, promotions, and stores connected by typed relationships
Config
Define your entity graph schema, constraints, and permission tiers in declarative YAML.
Ingest
Load entities and relationships from any source. Schema validation runs automatically.
Query
Ask causal questions. Agents execute deterministically and return receipts for every step.
Feedback
Record outcomes against predictions. The system calibrates edge weights and improves over time.
One engine, any domain
Define entity types and relationships in YAML. Cruxible Core handles the graph, the queries, and the receipts.
Sanctions & export control screening
Trace ownership chains, directorships, and offshore shell structures to surface hidden connections to sanctioned entities before transactions clear.
entity_types:SanctionedEntity:name: {type: string, indexed: true}country: {type: string}OffshoreCompany:Officer:relationships:- officer_of: Officer → OffshoreCompany- xref_officer: Officer → SanctionedEntity- sdn_connection: SanctionedEntity → SanctionedEntity
Start local, scale when ready
Cruxible Core runs on your machine — fast iteration, full privacy. When workloads outgrow a single machine, Cruxible Cloud adds managed compute, team access, and infrastructure you don't have to run.
| Cruxible Core Build with agents | Cruxible Cloud Scale with your team | |
|---|---|---|
| MCP server for agents | ||
| Knowledge graph engine | ||
| Deterministic execution | ||
| DAG-based receipts | ||
| Scope | Machine storage, thousands of entities | Managed storage, millions of entities |
| Persistent projects & snapshots | ||
| Async ingest & query jobs | ||
| Team access & RBAC | ||
| Audit logs & compliance export | ||
| API-compatible with local workflows | ||
| Price | Free, open source | Coming soon |
| View on GitHub | Join the waitlist |
MIT licensed, self-hosted
Cruxible Core runs locally. Read the source, audit the decision paths, extend the schema for your domain. Ship it in your own infrastructure.
View on GitHub$ pip install "cruxible-core[mcp]"
$ cruxible init --config config.yaml --data-dir ./data
Initialized graph: 0 entities, 0 relationships
$ cruxible ingest --mapping drugs --file data/drugs.csv
Loaded 1,247 entities, 3,891 relationships
$ cruxible query \
--query check_interactions --param drug_id=warfarin
✓ 12 interactions found — receipt: sha:b4e8f12
Solutions
Vertical-specific applications built on Cruxible Core.
Ready to build with Cruxible Core?
Start locally with the open-source engine, or join the Cruxible Cloud waitlist for self-serve deploys, managed infrastructure.