Developing with Agent Platform

To keep it super simple: when agents use integrations, they usually keep secrets and other sensitive data in their context window. In the future, agents will manage more and more user data - and all of that may end up unencrypted in context as well.

That's why we've developed Agent Platform as a thin abstraction over any of your llm calls - our Platform Gateway gates access to integration secrets, scopes permissions, does audit logging so you don't have to. All the agents see is a singular injected tool call that gives them a response (or tells them what is missing!) :)

Quickstart

1. Get your API key

Open the API Keys page in the dashboard and create a new key. Store it on your server:

export AP_API_KEY=ap_live_<your_key>

2. Create a session

A session holds conversation context across multiple requests, similar to a thread.

curl https://api.agentplatform.cloud/v1/sessions \
  -H "Authorization: Bearer $AP_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"title":"My first session"}'

{ "id": "sess_abc123", "status": "active" }

3. Send a prompt

Run a prompt inside your session. Set wait: true for a synchronous response, just like a regular model call.

curl https://api.agentplatform.cloud/v1/sessions/sess_abc123/runs \
  -H "Authorization: Bearer $AP_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "input": [
      {
        "role": "user",
        "content": [{ "type": "input_text", "text": "Summarize open issues in Linear." }]
      }
    ],
    "wait": true
  }'

{
  "id": "run_xyz789",
  "status": "completed",
  "output": [
    {
      "type": "message",
      "role": "assistant",
      "content": [{ "type": "output_text", "text": "There are 4 open issues..." }]
    }
  ]
}

That's it. Reuse the same session to continue the conversation, or create a new one for isolated tasks.

The full API spec is available at /v1/openapi.json.

Next steps