Local-first routing for AI agents

Why routing should stay close to your tools

Teams often combine hosted models, local Ollama, private endpoints, and several subscriptions. Each tool handles those connections differently, which makes outages and rate limits harder to recover from.

TEKIZ.AI puts those routes in one place, close to the machines and accounts you already control.

The problem with one bill and one marketplace

Hosted model marketplaces are convenient, but teams can give up three useful controls:

Privacy. Provider keys and traffic often move into another hosted service.
Control. Teams cannot always choose fallback order or whether local or cloud models run first.
Local models. Local Ollama and private Tailnet models are usually outside the natural path of a hosted marketplace, even though they are often the cheapest reliable fallback.

What local-first routing means

TEKIZ.AI runs on your machine, server, Tailnet host, or as a hosted service. It exposes one OpenAI-compatible endpoint. Your tools point at it. The router can:

Select models by task, speed, context size, file support, and your preferences.
Fail over across authorized providers when one is down, slow, or rate-limited, without switching models mid-stream.
Route local Ollama, Ollama Cloud through local Ollama, OpenAI, Anthropic, Gemini, NVIDIA NIM, private gateways, and OpenRouter BYOK routes.
Keep provider credentials local by default, while hosted TEKIZ.AI API keys manage account access, usage limits, and request history.
Record routing details so you can see which provider and model served a request.

For Codex and coding agents

Use one base URL and stable profiles such as tekizai/frontier while routing behind the scenes across healthy authorized providers.

For local Ollama fallback

Keep local models in the fallback chain for privacy, cost control, and continuity when cloud providers rate-limit.

For hosted teams

Hosted TEKIZ.AI adds team keys, usage limits, request history, health checks, and automatic fallback. Managed model access is included only when the selected plan says so.

A 30-second local start

python3 router.py --port 8790
export OPENAI_BASE_URL=http://localhost:8790/v1
export OPENAI_API_KEY=local-router
codex --model openai/gpt-4.1

Codex can now use your provider accounts and local models through the same address. Change the routes in TEKIZ.AI without configuring Codex again.

What the hosted plan includes

The local core is free and open source. Hosted plans include team keys, usage limits, provider health monitoring, dashboards, uptime checks, and automatic fallback. Bring authorized provider access unless your plan explicitly includes managed model access.

Next steps

Start with the hosted API quickstart, configure Codex, compare TEKIZ.AI vs OpenRouter, or estimate routing fit with the calculator.