Agentic governance
Your rules already exist.
Your agents don't know them.
Knowledge extracts your implicit rules, formalizes them in a living registry, and enforces them on every agent action — with end-to-end traceability.
scope: "finance" / ns: "risk"
scope: "finance" / ns: "pricing"
Current: 4.2% + requested +2% | Result: 6.2% > limit 5%
Purchase not recommended
The problem
You deploy AI agents. How do they know your rules?
- Business rules are scattered across multiple sources (PDFs, emails, wikis, prompts).
- There is no single source of truth your agents can query.
- Rule changes require manual updates across multiple agents.
- There is no audit trail linking agent decisions to the rules in force at the time.
How it works
See.·Formalize.·Apply.
Knowledge is a progression, not a monolith. Each level delivers standalone value — each step unlocks the next.
1 — See.
Knowledge surfaces implicit decisions from your repos, PRs, and docs.
Knowledge analyses your repos and generates typed drafts: candidate invariants, rules, and decisions. Nothing enters the registry without human validation.
- ✓Your own sources via stream API
- ✓Confidence score on each draft
- ✓Nothing enters the registry without human validation
- ✓Re-run in CI - the registry tracks the evolution of your practices
Tacit knowledge doesn't disappear — it enters the registry.
CI pipeline, triggered manually or at regular intervals. Zero developer friction.
2 — Formalize.
Validate the drafts. They become structured, versioned entries.
Four entry types organized by scopes and namespaces — fed by extraction or created directly in the web interface. Every rule is anchored to the decision that justifies it.
- Invariants
Absolute constraints
Non-negotiable limits. Blocking: if an agent violates one, the action is stopped. Some require human approval before any override.
e.g. "Maximum exposure of 5% per issuer."
- Rules
Active directives
Mandatory or advisory operational conventions. Versioned: when a rule evolves, the full history is preserved.
e.g. "Use EUR mid-swap curve for bond pricing."
- Decisions
Documented choices
Immutable records of significant choices — with context, reasoning, author, and timestamp. The rationale behind every rule.
e.g. "EUR mid-swap adopted as bond pricing benchmark — better liquidity for corporate spreads. Approved by head of pricing, March 2024."
- Overrides
Formal exceptions
A documented exception to a rule or invariant — with named approver, justification, conditions, and expiry. Not a workaround: a governed exception.
e.g. "30-day exception — exposure limit raised to 7% for TotalEnergies, approved by CIO."
- ✓Scopes & namespaces — organize rules at your organization's granularity
- ✓Versioned history — every change is traceable, every evolution is preserved
- ✓Queryable at any point in time — structured, not scattered
Rules are no longer in someone's head — they're in the registry, queryable, enforceable.
3 — Apply.
Your agents query Knowledge before acting
A minimal system prompt tells the agent to query Knowledge before acting. The rules themselves are fetched on demand — no prompt overload, no loss in long context.
- ✓REST API, compatible with any LLM and any framework
- ✓Native MCP : Claude Teams, Enterprise, Copilot Studio
- ✓Claude Code hooks : UserPromptSubmit, PostToolUse, Stop
- ✓GitHub Actions : compliance gate in CI
Governance is not a suggestion — it's a registry the agent queries before every significant action.
Use cases
For all your agents and copilots
Knowledge integrates into any context where LLMs operate on behalf of your organization — whether acting autonomously or assisting a human.
- 🏦
Finance & Trading
Exposure limits, compliance rules, pricing — every decision linked to the rule that enforced it.
- 💻
Coding agents
Claude Code and your CI/CD pipelines share the same architecture and security registry.
- ⚙️
CI/CD pipelines
No merge without compliance — the Verifier blocks non-compliant PRs automatically.
- 💬
Conversational assistants
Communication, confidentiality, and compliance rules enforced via MCP — no prompt to maintain.
- 🏥
Healthcare & Pharma
Medical invariants enforced on the copilot. Final decision stays human — Knowledge traces what was consulted.
- ⚖️
Legal & Contractual
Mandatory clauses, liability caps, approval levels — compliance traced at every interaction.
Why Knowledge
Beyond the system prompt
A system prompt configures one agent. Knowledge enforces rules across all your agents — with traceability, versioning, and accountability.
Updates
System Prompt
Manual, per instance
✦ Knowledge
One rule changed, all agents updated
Enforcement
System Prompt
Forgettable suggestion
✦ Knowledge
Injected hook — LLM-independent
Contextual memory
System Prompt
Forgotten in long sessions
✦ Knowledge
Re-injected at every prompt
Exceptions
System Prompt
No formal mechanism
✦ Knowledge
Override documented, approved, expirable
Compliance proof
System Prompt
Declared intent
✦ Knowledge
Traced and auditable execution
| System Prompt | ✦ Knowledge | |
|---|---|---|
| Updates | Manual, per instance | One rule changed, all agents updated |
| Enforcement | Forgettable suggestion | Injected hook — LLM-independent |
| Contextual memory | Forgotten in long sessions | Re-injected at every prompt |
| Exceptions | No formal mechanism | Override documented, approved, expirable |
| Compliance proof | Declared intent | Traced and auditable execution |
Compliance
Designed for the European AI Act
Knowledge natively addresses the requirements of high-risk AI systems — not as an added feature, but by design.
Risk management
Codified invariants. Distill detects obsolescence.
Technical documentation
Immutable decisions with context and reasoning.
Record-keeping
Automatic logging of every agent interaction.
Transparency
Versioned registry, consultable at any point in time.
Human oversight
Approvals, formal overrides, roles and permissions.
Robustness
Enforcement via hooks — LLM-independent.
See how your agents use your rules
Coming soon — an agentic compliance dashboard: followed/diverged rate, never-queried rules, Distill obsolescence signals.
Pricing
Connect your agent on day 1.
The registry is accessible from the free plan. Extraction is a feature, not a tier.
Starter
Free
The registry from day one
- ✓MCP server + REST API — connect any agent
- ✓resolve(), check(), query() — full runtime access
- ✓1 scope / 3 namespaces
- ✓Invariants, Rules, Decisions, Overrides
- ✓Distill extraction — one-shot
- ✓3 users
- ✓Community support (GitHub)
Team
€299/month
Scale across your organization
Everything in Starter, plus:
- ✓Unlimited scopes & namespaces
- ✓Dependency graph and typed relations
- ✓Overrides & approval workflow
- ✓Unlimited history and event timeline
- ✓CI/CD Verifier — compliance gate
- ✓Distill re-run — obsolescence detection
- ✓Priority email support
Enterprise
Custom
End-to-end governance at scale
Everything in Team, plus:
- ✓Multi-tenant (holding → subsidiaries)
- ✓SSO & SCIM
- ✓On-premise deployment
- ✓Guaranteed SLA
- ✓AI Act compliance support and third-party audit
- ✓Custom Distill connectors (Notion, Confluence, SharePoint)
Ready to start?
Your rules already exist. It's time your agents knew them.
Start with Distill — make the implicit visible. Then formalize in a living registry. Then enforce on every agent action.