Assistiv Docs

[ REFERENCE ]

Agents

apiv0.1.0sdk@assistiv/sdk@0.2.0

Agents are callable as models. Every agent on your platform is a model named assistiv-ai/<agent-slug> on the OpenAI-compatible endpoints — same request shape, same streaming, same governance as plain inference.

Provisioning is invite-only

Agents are set up per platform during white-glove onboarding. If you don't have one yet, talk to the founders and we'll configure it with you — model, instructions, tools, and skills included.

Call an agent

Use the end-user's sk-eu_*key and the agent's model name. Works on /v1/chat/completions and /v1/responses.

bash
curl -X POST https://api.assistiv.ai/v1/chat/completions \
  -H "Authorization: Bearer $END_USER_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "assistiv-ai/support-bot",
    "messages": [{"role":"user","content":"Summarize ticket #4812 and draft a reply"}],
    "thread_id": "thread_4812",
    "stream": true
  }'

What you get on every run

Multi-turn sessions

Pass thread_id to continue a conversation. Agent state persists server-side per thread — no message replay from your app.

Tools and skills

Agents use their configured tools, hosted MCP apps, and skills during a run. You can also pass request-level tools exactly as with plain inference.

Streaming

stream: true yields server-sent events, matching the OpenAI streaming format.

Governance built in

Budget pre-flight before the run, per-token and per-tool debits during it, and one audit-ready log row after it. A 402 means the wallet or budget gate fired — same semantics as any inference call.

Scope and limits

Agents are API-invoked: a run starts when you call the endpoint, and there is no scheduler or background trigger. Session memory is per-thread; agents do not learn across threads.

Next steps