OpenInfer

Build the future of creative AI.

Your key is your OpenAI-compatible sandbox — text and vision inference, ready to build on. Paste it to start.

No key yet? Ask an organizer, or email hackathon@openinfer.io.

Run it live. Then build.

OpenInfer is your OpenAI-compatible sandbox — a distributed inference engine for text and vision. Run each call live below, copy the code, and ship something that moves the field forward.

Three calls — run each one live

Generate text

Send a prompt, stream the response back token by token. The core call behind chat, content, and agent features — anything text in your app.

Run a prompt to stream the response here.
Idle 
…
Next: understand images

Understand images

Send an image and a question; get a description back. Build visual tools that empower anyone to create — not just designers.

Image
Add an image and run to get a description.
Idle 
…
Next: list models

List models

See which models your key can use. Build on @oi/beta.

No prompt needed — this fetches the live list of models your key can access through the engine.

Run to see the models your key can use.
Idle 
…
You're set — start building

That's the API. Now go build.

Copy the snippets, wire them into your app, and build AI that designs, creates, and inspires. Full reference is below.

Notes for the maker.

Request fields — POST /v1/responses

model req
Model ID. Use @oi/beta.
stream req
Must be true for now — the server replies with Server-Sent Events.
input req
A plain string, or an array of message objects (with input_text / input_image parts).
instructions
System prompt override; takes precedence over system/developer messages.
temperature
Sampling temperature for this request.
max_output_tokens
Maximum generated tokens.
store
Defaults to true. Stores the response for retrieval or continuation.
previous_response_id
Continue from a stored response. Must reference the current head.
tools / tool_choice
Function tools in the flat Responses format; support is limited.

Streaming events (SSE)

  • 01response.created
  • 02response.in_progress
  • 03response.output_text.delta — consume for live display
  • 04response.output_text.done
  • 05response.completed — the final object

Going further — where a database helps

Conversation history

Persist turns and continue a stored response with previous_response_id.

Agent memory

Facts, summaries, and embeddings your agent recalls across sessions.

Auth & sessions

Users, API keys, and session tokens for your own app.

Output & image cache

Cache generations and uploads so you never recompute a piece.

Config storage

Per-user settings. Building on OpenClaw? Point it at OpenInfer and store its state in your DB.

Logs & analytics

Request and usage logs to debug and measure your hackathon project.

Sponsor · MongoDB

Give your agent a memory with MongoDB.

Agentic systems need state that outlives a single request. MongoDB Atlas stores conversation history, agent memory, and tool results as flexible documents — and its built-in vector search powers semantic recall and RAG, so your agent retrieves the right context instead of re-deriving it every turn.

Get started with the MongoDB Agentic Memory & Context Engineering hackathon guide — setup, Atlas credits, and resources.