# IFM Documentation

> Docs for foundation models from the Institute of Foundation Models (IFM) at MBZUAI — K2 Horizon, Jais 3, and Jais ASR — served through an OpenAI-compatible API at https://api.ifm.ai/v1.

Every page below is plain Markdown. Each page is also served as static HTML at `https://docs.ifm.ai/<slug>/`. The full documentation in one file: [llms-full.txt](https://docs.ifm.ai/llms-full.txt).

## Get Started

- [Overview](https://docs.ifm.ai/docs/overview.md): Model portfolio, featured models, and where to start.
- [Quickstart](https://docs.ifm.ai/docs/quickstart.md): Create an API key, set the base URL, and make your first chat completion.

## Models

- [Model overview](https://docs.ifm.ai/docs/model-catalog.md): Every IFM model with context length, languages, and hosted-API availability.
- [K2 Horizon](https://docs.ifm.ai/docs/introduction.md): Frontier agentic reasoning series; six open-weight sizes, 375B served on the API.
- [Jais 3](https://docs.ifm.ai/docs/jais-3.md): Arabic-specialized agentic model on the K2 Horizon architecture; IFM/Jais-3-MoVA-36B-A4B on the API.
- [Jais ASR](https://docs.ifm.ai/docs/jais-asr.md): Arabic speech recognition with dialect ID, code-switching, and diacritization.
- [K2 V2](https://docs.ifm.ai/docs/k2-v2.md): Fully-open 70B research models (deprecated on the API; weights remain).
- [Jais 2](https://docs.ifm.ai/docs/jais-2.md): Arabic–English chat models, 70B and 8B (weights only).

## Capabilities

- [Reasoning](https://docs.ifm.ai/docs/reasoning.md): reasoning_effort levels, reasoning_content, and per-family support.
- [Function calling](https://docs.ifm.ai/docs/custom-function-calling.md): OpenAI-schema tools, the tool loop, and tool_choice.
- [Streaming](https://docs.ifm.ai/docs/streaming.md): SSE deltas for content, reasoning, and tool calls.
- [Structured output](https://docs.ifm.ai/docs/structured-output.md): JSON mode and JSON Schema via response_format.
- [Multi-turn conversations](https://docs.ifm.ai/docs/multi-turn.md): Replaying assistant turns and reasoning_content verbatim.
- [Session routing](https://docs.ifm.ai/docs/session-routing.md): X-Session-ID for KV-cache affinity.

## API Reference

- [Chat completions API](https://docs.ifm.ai/docs/chat-completions.md): POST /v1/chat/completions — parameters and response schema.
- [Transcriptions API](https://docs.ifm.ai/docs/transcription.md): POST /v1/audio/transcriptions — multipart request, language codes, response, formats.
- [Limits](https://docs.ifm.ai/docs/limits.md): Daily token cap per key and rate-limit headers.
- [Errors](https://docs.ifm.ai/docs/errors.md): Status codes, error types, and retry guidance.

## Deploy

- [Deploy K2 Horizon](https://docs.ifm.ai/docs/deployment.md): Self-host with SGLang or vLLM.
- [Deploy Jais 3](https://docs.ifm.ai/docs/deploy-jais-3.md): Self-host with vLLM.

## Harness

- [OpenCode](https://docs.ifm.ai/docs/opencode.md): Configure OpenCode with an IFM provider.
- [OpenClaw](https://docs.ifm.ai/docs/openclaw.md): Register IFM as an OpenClaw model provider.
- [Pi](https://docs.ifm.ai/docs/pi.md): Configure the Pi coding agent for IFM.
- [Hermes](https://docs.ifm.ai/docs/hermes.md): IFM provider plugin and config for Hermes Agent.

## Lifecycle

- [Migration to K2 Horizon](https://docs.ifm.ai/docs/migration.md): Moving from K2 V2: model name, context, reasoning replay, tool formats.
- [Deprecation](https://docs.ifm.ai/docs/deprecation.md): Retired models and base URLs with replacements and dates.

## Optional

- [MBZUAI Privacy Policy](https://mbzuai.ac.ae/privacy-policy)
- [IFM on Hugging Face](https://huggingface.co/IFM)
- [IFM Platform (API keys)](https://platform.ifm.ai/api-keys)