Local Ollama stays local
Run TEKIZ.AI beside your local Ollama runtime and point Codex, OpenClaw, Cursor, Aider, Continue, OpenHands, or any OpenAI SDK client at one local base URL.
Ollama routing for agents
Give your tools one OpenAI-compatible address for local Ollama, Ollama Cloud, and any other models you connect. TEKIZ.AI chooses an available model based on cost, fit, and reliability.
Run TEKIZ.AI beside your local Ollama runtime and point Codex, OpenClaw, Cursor, Aider, Continue, OpenHands, or any OpenAI SDK client at one local base URL.
Use Ollama Cloud models through the same setup, with the same health history and backup order as your other providers and private models.
When a cloud key or model hits a retryable rate limit, TEKIZ.AI can try another key, provider, or local Ollama model.
export OLLAMA_HOST=http://127.0.0.1:11434
python3 router.py --port 8790
export OPENAI_BASE_URL=http://127.0.0.1:8790/v1
export OPENAI_API_KEY=local-router
curl "$OPENAI_BASE_URL/chat/completions" \
-H "Authorization: Bearer $OPENAI_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "tekizai/frontier",
"messages": [{"role": "user", "content": "Route this request to an approved local or cloud model"}]
}'
Add several OpenAI, Anthropic, Gemini, or compatible keys. TEKIZ.AI Router spreads requests across keys that are working and pauses only the one that is rate-limited.
Images, audio, video, and documents are sent only to models that can handle that kind of input.
Use a profile such as tekizai/frontier or name an
Ollama model directly. TEKIZ.AI checks availability and falls
back according to your settings.
Direct Ollama is enough for one machine and model. TEKIZ.AI adds cloud backup, 429 failover, multiple API keys, file compatibility checks, and private Tailnet access while keeping Ollama available as your local option.
Create an API key, read the hosted/local quickstart, see the self-hosted path, or compare OpenRouter.
If your company approves only Ollama, keep that boundary. TEKIZ.AI can reduce unnecessary token processing and make better use of existing compute.