Three controls at the boundary, one platform behind them.
Chrysent is an integrated, self-hosted platform for governing every AI interaction and agent action — from prompt to tool execution. Gateway, Touchstone, and Bastion sit at the request, content, and execution boundaries. Axis Code, Axis Knowledge Base, Deep Research, OPI Forge, and Chrysent Sandbox connect code, knowledge, research, delivery, and sandbox controls to the same governance and evidence model.
Architecture
Static diagram of the request, policy, context, and execution flow. Edge labels describe what crosses each boundary. Status is indicated by line style and text (not color alone).
Architectural invariants
The platform commits to a small set of architectural invariants. They govern credential handling, the relationship between enforcement and effect, the separation between policy and execution, and how configuration rolls out. Buyers can rely on these; technical evaluators can challenge them.
Native contracts first
OpenAI Responses, Anthropic Messages, and Chat Completions are passed through to compatible provider endpoints by default. Translation is explicit, not automatic, and its limits must be visible.
Customer policy is data
Customer safety customization is preset selection, rule parameters, extraction mappings, scoped exclusions, category thresholds, and configured dispositions — never executable code, never network or filesystem access.
Enforcement precedes effect
A tool call proposal MUST NOT authorize an action. Bastion is configured as a pre-tool-execution hook in the agent harness: it evaluates each proposed call against customizable policy and returns allow, deny, or human-review decisions. It never executes tools, and no call may execute before its policy decision is recorded.
The request data plane is stateless
Request and stream state is transient. Durable configuration, identity, credentials, evidence, and search live outside request-serving replicas.
Configuration and binaries rollout independently
Policy and software updates are rolled out separately. Operations pin one component snapshot per Gateway, Bastion, or Touchstone operation. The end-to-end interaction records the complete version set used.
BYOK is the production customer path
Production inference is BYOK. Chrysent does not sell inference.
Inference providers use an allowlist
Gateway routes only to enrolled provider endpoints. Provider origins are operator-owned: tenant input cannot alter scheme, authority, or port. Credentials never cross origins through redirects.
Gateway terminates credentials
Gateway terminates Chrysent and client authentication and never forwards either upstream. It injects only the route-resolved provider credential after final destination validation and never returns provider credentials to clients.
Availability
Status reflects current capability and access scope. The same vocabulary is used on every product page.
| Product | Status | Role | Notes |
|---|---|---|---|
| Chrysent AI Gateway | Private beta | Governance | Available in private beta. Production deployment requires a signed evaluation agreement. |
| Chrysent Bastion | Private beta | Defense | Available in private beta. Policy engine, tool-call mediation, and audit trace are exposed in the beta. |
| Chrysent Touchstone | Private beta | Defense | Available in private beta. Prompt-injection detection, redaction, and policy routing are exposed in the beta. |
| Chrysent Axis Code | Private beta | Engineering | Code graph indexing and retrieval are running in a closed beta. Capacity is limited. |
| Chrysent Axis Knowledge Base | Design-partner preview | Memory | Available to design partners. Connector ingestion and permission-aware retrieval are exposed in the preview. |
| Chrysent Deep Research | Private beta | Research | Long-horizon research workflows are available in private beta. Outputs require human review. |
| Chrysent Sandbox | Design-partner preview | Isolation | Policy-controlled sandbox access and isolation capabilities are available to design partners. |
| OPI Forge | Private beta | Engineering | Specification-driven agentic software delivery is available in private beta. |
Start with your problem
Most design partners start with one product and expand to the full wedge within 90 days.
Cost & governance
Multi-vendor model traffic, attributed spend, audit, and budget controls in one control layer.
Private betaPrompt safety
Detect prompt injection, jailbreaks, and prohibited content; redact sensitive data before it reaches the model.
Private betaAgent execution
Inspect agent tool calls before they reach files, machines, or MCP servers; explain or conceal rejections.
Private beta
Role-based pathways
What different stakeholders gain from the platform.
CISO
Problem: How do I audit AI usage?
Gain: Session recording + replay, encrypted at rest and in transit. Every prompt, response, and tool call is captured for review.
See the relevant product →CFO
Problem: What's this costing us?
Gain: Per-team, per-project cost attribution with budget alerts and circuit breakers. No more "who spent $50K on tokens?"
See the relevant product →Legal / Compliance
Problem: Are we EU AI Act ready?
Gain: PII redaction before egress, data residency controls, and audit trails aligned with the controls auditors expect.
See the relevant product →VP Engineering
Problem: How do we manage keys?
Gain: Key pooling, circuit breakers, and model routing. Single ingress point across providers with no vendor lock-in.
See the relevant product →CTO
Problem: Are we locked into one vendor?
Gain: Vendor-agnostic, any model, native passthrough mode. OpenAI Responses, Anthropic Messages, and Chat Completions route as-is.
See the relevant product →
Deployment and data handling
Chrysent sells self-hosted software only. Software and retained product data remain in customer-owned or customer-controlled infrastructure. Customer-configured external model providers receive provider-bound request data under the customer's provider agreement. This model supports data sovereignty and residency requirements. Deployment, retention, and customer-configured provider boundaries are described on the Data handling & deployment page. Security controls and disclosure are on the Security & trust page.