Skip to main content
Agents API is currently in early beta. The API may change before general availability.
The agent operates in sandboxed environments where they can execute code, access the web, manage files, and complete multistep tasks, all streaming results back in real-time. The API is powered by Raccoon AI’s Agents SDK.

Key Concepts

Sessions

Persistent workspaces identified by a UUID. The agent maintains context across turns: files, conversation history, credentials and processes persist.

Turns

Single request-response cycles. Each turn builds on previous context within the same session. The agent sees prior work and continues incrementally.

Modes

Control how the agent works:
ModeUse case
autoGeneral-purpose (default)
codeSoftware development
webWeb apps with live preview
deepsearchResearch and analysis
slidesPresentations
planPlanning before execution
chatConversation only

How It Works

1

Start a turn

POST /chat/new with your prompt, mode, and session ID.
2

Stream results

Connect to wss://api.raccoonai.tech/ws/chat/{taskId} to receive events as the agent executes.
3

Continue or branch

Reuse the same chat_id to continue the session, or generate a new UUID to start fresh.

Endpoints

MethodEndpointDescription
POST/chat/newStart a turn
DELETE/message/{messageId}/abortStop a running turn
GET/chatsList sessions
GET/chat/{chatId}Get session details
PUT/chat/{chatId}/titleRename session
DELETE/chat/{chatId}Delete session
PUT/chats/media/uploadUpload files
WSS/ws/chat/{taskId}Real-time events

Rate Limits

Rate limits are enforced based on your plan. The X-Rate-Limit-Max-Concurrent header indicates the maximum turns you can run simultaneously. Exceeding limits returns 429 with an X-Rate-Limit-Retry-After header. For custom limits and pricing, contact sales.