Build AI agents your compliance team will love. Full provenance tracking, policy enforcement, and decision traces for every action.
import { ContextGraph } from '@contextgraph/sdk';
const cg = new ContextGraph();
// Every claim requires provenance
await cg.addClaim({
entity: 'report_q4',
attribute: 'status',
value: 'verified',
source: { type: 'analyst', id: 'user_123' },
confidence: 0.95,
validFrom: now()
});
// Policy-evaluated decisions
const decision = await cg.proposeDecision({
type: 'publish_report',
target: 'external',
evidence: [claim.id]
});
// decision.status: 'approved' | 'needs_review'AI agents are everywhere. Audit trails are nowhere.
Agent made a bad call? Good luck figuring out why. No trace, no evidence, no accountability.
Auditor asks "show me every decision this agent made in Q3." You: *sweating*
Agent has access to everything. Policies exist in a wiki somewhere. Hope it reads that wiki.
"What did we know when we made that decision?" Nobody knows. Data is overwritten daily.
A governance substrate that makes auditability automatic.
Every piece of data requires a source. No orphan facts. Every claim is traceable to its origin.
source: { type: 'api', id: 'crm_v2', timestamp: now() }
All data is temporally qualified. Ask "what did we know on March 15?" and get an answer.
validFrom: '2024-01-01', validUntil: '2024-12-31'
Every decision is recorded before execution. Full lifecycle: proposed, approved, executed, completed.
status: 'proposed' | 'approved' | 'executed'
Deny-first policy evaluation. No action proceeds without explicit authorization.
effect: 'deny', unless: { role: 'admin' }
High-risk decisions queue for human review. Approval workflows built-in.
requiresApproval: risk.level >= 'HIGH'
Generate audit reports on demand. SOC 2, GDPR, HIPAA evidence packages.
reporter.generateSOC2({ period: 'Q4' })
Four planes. Zero black boxes.
"No agent action is valid unless it is context-justified, policy-evaluated, and decision-traceable."
Interactive demos showing real governance scenarios.
Watch how a high-risk action is caught by policy, queued for human review, and then executed after approval.
Query what the system knew at any point in time. Essential for "why did we make that decision?" questions.
Generate audit-ready reports with one command. All decisions, evidence, and provenance included.
All the power of ContextGraph OS, none of the infrastructure hassle.
Monitor every agent decision, policy evaluation, and approval workflow in one beautiful interface. Built for teams who need visibility and control.
Full isolation with row-level security
SAML 2.0, OIDC, Okta, Azure AD
HMAC-signed events, full REST API
Docker, Kubernetes, Helm charts
Built for teams who can't afford black-box AI.
Finance, healthcare, legal. When the auditor asks, you'll have answers.
Building internal agents? Make them audit-ready from day one.
Building agent platforms for others? Give them governance for free.
Start free. Scale with confidence.
For developers and open source projects
For teams building AI-powered products
For regulated industries & large deployments
| Feature | Open Source | Team | Enterprise |
|---|---|---|---|
| Core SDK & Packages | 22 packages | 22 packages | 22 packages |
| Hosted Dashboard | — | ✓ Included | ✓ Included |
| Storage Backend | SQLite | PostgreSQL | PostgreSQL |
| Agents | Unlimited | Up to 10 | Unlimited |
| Audit Retention | Self-managed | 90 days | 1 year+ |
| SSO / SAML | — | ✓ Included | ✓ Included |
| Compliance Reports | — | Basic | SOC2, HIPAA, GDPR |
| Support SLA | Community | 24hr response | 4hr response |
| License | AGPL-3.0 | Commercial | Enterprise |
Common questions about ContextGraph OS.
No. ContextGraph OS is a governance substrate that sits between your agent framework (LangChain, AutoGen, etc.) and actual execution. It doesn't replace your agents; it makes them auditable.
Minimal. Policy evaluation is in-memory and typically adds <10ms. For high-risk actions that require human approval, yes there's latency - but that's the point. You want someone to review before the agent deletes your production database.
Currently SQLite (embedded) and an in-memory adapter for testing. PostgreSQL adapter is on the roadmap. The storage layer is abstracted, so adding new backends is straightforward.
Yes. The SDK provides a simple API. Your agent submits "action intents" instead of executing directly. ContextGraph evaluates policies, records the decision, and either allows execution or queues for review.
Dual-licensed: AGPL-3.0 for open source use, and a commercial license for proprietary applications. If you're building internal tools or open source, AGPL is free. If you're building a SaaS product, you'll need a commercial license.
ContextGraph OS is unique in its focus on governance as a first-class concern. Other tools might have logging or monitoring. We have provenance, temporal data, policy enforcement, decision traces, and compliance reporting baked in at every layer.
Start with open source, upgrade to Cloud when you need more. Build trust with every decision.