Sogni: Learn logo

💡Sogni Intelligence Introduction

Sogni Intelligence is an OpenAI-compatible LLM and agent API running on the Sogni Supernet. It supports normal chat, streaming, vision input, custom function calling, and Sogni media tools for image, video, and music generation.

The main synchronous entry point is:

POST https://api.sogni.ai/v1/chat/completions

Use /v1/chat/completions when an LLM should interpret a user request, decide whether tools are needed, and return a final assistant response. Use /v1/chat/runs when you want that same model-selected hosted tool behavior with durable progress, replayable events, recovery, and cancellation. Use /v1/creative-agent/workflows when your application already knows the exact media workflow to run.

#Current Model IDs

Use GET /v1/models for the live list. Current public Sogni Intelligence models include:

Model ID Use
qwen3.6-35b-a3b-gguf-iq4xs Default model for chat, reasoning, coding, vision, and tool calling.
qwen3.5-35b-a3b-abliterated-gguf-q4km Alternative Qwen 3.5 model.

#What It Supports

  • OpenAI-compatible chat completions
  • Server-Sent Events streaming
  • Multi-turn conversations
  • developer, system, user, assistant, and tool messages
  • Inline base64 PNG/JPEG vision input
  • Custom OpenAI-style function tools
  • Default creative media tools with media indexing across tool rounds
  • Optional server-side tool execution with durable workflow records
  • Synchronous composition tools for prompt enhancement, scripts/storyboards, lyrics, and instrumental structures
  • Synchronous workflow planners (compose_workflow, compose_workflow_template) that emit a ready-to-submit durable workflow plan from a brief
  • Durable hosted chat runs with persisted LLM/tool rounds, event replay, SSE progress, recovery, and cancellation
  • Durable creative workflow recovery, event replay, SSE progress, resume, and cancellation

#API Guides

#Agent Runtime Skill

If you want an existing agent runtime to use Sogni without building directly against the REST API, install the public Sogni Creative Agent Skill. It provides the sogni-agent CLI plus a SKILL.md adapter for Claude Code, OpenClaw, Hermes Agent, Manus, and other skill-based agents. The CLI can call /v1/chat/completions with --api-chat or durable /v1/creative-agent/workflows with --api-workflow, while direct media workflows continue to use local --ref, --ref-audio, and --ref-video flags.

#Tool Families

By default, Sogni injects the creative-tools family used by Sogni Chat. It includes image, video, music, image/video analysis, metadata extraction, and media indexing across tool rounds. The legacy sogni_tools: "rich" alias is accepted as the same creative media surface.

The default creative-tools surface includes the full creative media tool surface plus synchronous composition tools: enhance_prompt for model-ready prompt preparation, compose_script for scripts/storyboards/campaign beats/social shorts, compose_lyrics for vocal songs, and compose_instrumental for instrumental structures. Set sogni_tools: "creative-agent" when you also want workflow control, asset-manifest tools, and the synchronous workflow planners compose_workflow and compose_workflow_template. Set sogni_tools: false / "none" for plain text or coding-only clients that do not need Sogni tools.

#Quick Test

curl https://api.sogni.ai/v1/chat/completions \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "qwen3.6-35b-a3b-gguf-iq4xs",
    "messages": [
      {"role": "user", "content": "Hello. What can you help me build?"}
    ],
    "sogni_tools": false
  }'

See Chat Completions for request fields, response behavior, tool calling, replay records, and OpenAI SDK compatibility. See Durable Chat Runs for /v1/chat/runs state, events, cancellation, and recovery. See Creative-Agent Workflows for durable workflow orchestration.

Last updated 2026-04-23