Multi-model routing

TEKIZ.AI Fusion

Split independent tasks across available models, check their work, and return one OpenAI-compatible response.

Set up Fusion Run locally

Fusion is available on Pro and Max when worker and review models are online. A verified account can use the bounded proof window when it is offered.

Checking Fusion availability...

Community provider alpha Fusion Swarm roadmap Live status

How it works

Run independent tasks in parallel.

Fusion sends numbered or bulleted tasks to faster cloud or local models when they can be checked independently. A separate model reviews the work and returns the final answer. Offline or unavailable routes are skipped.

  • Research: compare sources and tradeoffs.
  • Implementation: delegate bounded changes while a stronger route reviews them.
  • Agents: receive one normal completion, with no new SDK. 
curl "https://api.tekiz.ai/v1/chat/completions" \
  -H "Authorization: Bearer sk_tekiz_..." \
  -H "Content-Type: application/json" \
  -d '{
    "model": "tekizai/fusion",
    "messages": [
      {"role": "user", "content": "1. Compare the launch options against cost and delivery risk.\n2. Identify the two material implementation risks.\n3. Recommend one option with a concise rationale."}
    ]
  }'

For existing agents

Give Frontier explicit work to send through Fusion.

Gateway-style clients can attach tekizai:fusion without forwarding unknown tool markers downstream. Set tool_choice to required for explicit invocation, and use a short list when work can run independently. 

{
  "model": "tekizai/frontier",
  "messages": [
    {"role": "user", "content": "1. Summarize the strongest case for this launch.\n2. Summarize the strongest case against it.\n3. Recommend a next step with the decision criteria."}
  ],
  "tools": [{"type": "tekizai:fusion"}],
  "tool_choice": "required"
}

Access and privacy

Use only models you are allowed to use.

Hosted requests use TEKIZ.AI cloud models or approved Community nodes. Personal provider accounts are never shared with other customers.

  • Hosted: Fusion runs only when both worker and review models are available.
  • Local: use the same model alias with providers and local models you authorize.
  • Privacy: logs keep request timing and route health, never prompts, model output, API keys, OAuth tokens, or provider credentials.