Send one request through Praesidia, then read the decision it left behind.
Start with a free workspace and one harmless request. Praesidia identifies the caller, applies your rule, and writes the decision into a signed, chained audit record you can export and verify offline.
The TypeScript and Python SDKs are not on npm or PyPI yet; design partners receive them directly. Everything on this page works over HTTP without them. No credit card required.
curl --fail-with-body "${PRAESIDIA_GATEWAY_ORIGIN}/openai/v1/chat/completions" \
-H "Authorization: Bearer ${PRAESIDIA_APPLICATION_KEY}" \
-H 'Content-Type: application/json' \
--data "{\"model\":\"${PRAESIDIA_MODEL}\",\"messages\":[{\"role\":\"user\",\"content\":\"Say hello in one sentence.\"}]}"From the model gateway quickstart. Set the three variables to your gateway origin, your application key and a model enabled for your route; no SDK is needed.
From a key to evidence someone else can check
Each step links the guide that covers it. Start in a test workspace with a request whose effect you understand.
- 1
Create a key. In your workspace, create an application and its application key for the model gateway. A runtime credential for governed agent connections comes from your deployment operator; signup does not issue one on its own. Keys and routes
- 2
Wrap the action. Register the agent and the tool it calls, then connect them so the call runs through a route Praesidia governs, with a rule on it. Connections · Policies
- 3
Execute a request. Send one request the rule should allow and one it should stop, like the command above. Connection quickstart
- 4
See the decision. Find both requests in the audit trail: who asked, what for, which rule decided, and the outcome. Action to evidence
- 5
Verify the evidence. Export a signed bundle and check it offline with the verifier, which we hand to you directly. What the verifier checks
New policies start in observe mode: the decision is recorded and the request goes through. Denials apply once a policy is set to enforce.
What you can use today, and how to get the rest
- Frameworks and runtimes
- LangGraph, CrewAI, OpenAI Agents SDK, Google ADK and more, each with the recipe we have validated and what it covers: frameworks. For coding agents, workflow tools and MCP clients, see the runtime catalog.
- SDKs
- TypeScript and Python SDKs exist and are not yet published to npm or PyPI. Design partners receive them directly, and runtime installation covers installing from a provided release. Ask for access.
- The offline verifier
- Handed to customers and their auditors directly; it is not a public download yet. Request it.
- Plans
- The signed audit trail, rule-based guardrails, API keys and bring-your-own-key model configuration are in every plan, including the free one. Protected-action evidence is on Advanced and is enabled per organization; human approvals are Enterprise. Compare plans.
Read next: choose your first outcome · all docs · the evidence · runtime security
One request, one rule, one record
Create a workspace and follow the quickstart with a harmless test request.