Intelligence as a service
An agentic AI backend, behind one HTTP call.
Amy runs the reason–act loop for you: your request goes in, and tokens, tool calls, delegated specialists and token usage stream back. 14 endpoints, 13 event types, one API key.
Prices in ZAR from R0/month. See plans.
curl -N https://api.askamy.co.za/v1/agents/amy/runs \
-H "X-Api-Key: $AMY_API_KEY" \
-H "Content-Type: application/json" \
-d '{"message": "Summarise this quarter'"'"'s support tickets by theme."}'event: run_started
data: {"runId":"01936c2f-…"}
event: token
data: {"text":"Three themes dominate"}
event: tool_call
data: {"callId":"c1","toolName":"recall","argsJson":"{…}"}
event: usage
data: {"inputTokens":1841,"outputTokens":612,"totalTokens":2453}
event: run_completed
data: {"runId":"01936c2f-…","status":"completed"}One endpoint, a whole agent
Post a message. Amy plans, calls tools, delegates to specialists and comes back with an answer. You do not orchestrate a loop, manage a context window, or hold provider SDKs.
Streamed, awaited or queued
The same run, three shapes: stream it token by token over SSE, await one JSON result, or submit it and poll. Pick per call — no separate integration.
Your tools, not ours
Register a Model Context Protocol server and the agent can call your systems. A tool marked for approval pauses the run until a human decides, before the tool runs.
Metered and auditable
Every run persists its messages, tool calls and token usage. The run’s own total covers everything it did — delegated specialists included — so you can attribute spend to your own customers.
The docs are the product tour.
Every endpoint in the reference is generated from the API's own OpenAPI document, so the reference cannot drift from what the service actually does.