One control plane, two ways to start
Start with your own agent or co-build with us. Either way, you get the same policy checks, human gates, and evidence trail.
Bring your own agent
Connect the agent stack you already run to Verdict at the point where high-risk actions execute. Keep your models, tools, and orchestration; add a control plane that decides before side effects.
- Synchronous policy check before every side-effecting action
- TypeScript and Python SDKs with AutonomousEnforcer execution loop
- Shadow mode, then hard enforcement, without redeploying the agent
Verdict-built FinTech agents
We ship production FinTech workflow agents end-to-end on the same control plane, then operate that layer with you as you scale.
- Fraud operations, payment disputes, and card-issuing workloads on one enforcement surface
- Your team operates with full control day-to-day; core workflow agent IP remains with Verdict under agreed terms
- Workflow policy packs, compliance harness, and signed evidence built in from day one
Governance primitives
Both routes use the same core controls: one policy model, one approval flow, and one evidence trail. Need implementation detail? See How it works.
Six primitives · control plane → evidence → lifecycle
Agent governance control plane
A synchronous policy enforcement gateway on the tool-call path: each instrumented action is evaluated before execution, with server-side state held outside the model’s direct control in the default architecture.
Three-outcome policy model
ALLOW, DENY, or REQUIRE_CHANGES with structured repair hints, so agents recover inside policy with a clear next step.
Compliance harness
Consent, disclosure, approval, secure input, and temporal steps are first-class obligations wired into policy evaluation and rendered through the harness.
PII tokenization (VT_)
Sensitive fields become opaque tokens before the LLM sees them; the gateway resolves values server-side when tools execute.
Cryptographic evidence plane
Hash-chained, Ed25519-signed decision records with tamper-evident artifacts for audit, model risk, and regulatory review workflows.
Policy hot-reload & replay
Policy-as-code (Rego/OPA bundles) updates in seconds without redeploying agents; replay new rules on historical tool-call traces before go-live.
Start with one workflow you can defend in production
In a working session we align on where actions are enforced, how rules map to your SOPs, and what evidence model-risk and audit need, so engineering isn’t guessing.