Providers & models

Tachyon is bring-your-own-key across 130+ providers. How to store credentials, pick a model, and switch mid-session.

How BYOK works

Tachyon talks to your provider directly using credentials you supply. There is no Tachyon account, no hosted relay, and no per-seat subscription — your prompts and code go from your machine to the provider you chose.

The provider and model catalogue comes from models.dev, which is why the list is 130+ entries long and stays current without a Tachyon release. Model ids follow provider:model, e.g. anthropic:claude-sonnet-5.

Storing a key

bash

tachyon login anthropic

You are prompted for the key, which is written to your user config directory — not the repository — so it is never at risk of being committed. Run login once per provider you intend to use; keys for several providers can coexist.

Choosing a model

With no argument, browse the catalogue interactively:

bash

tachyon model

With an id, set the default outright:

bash

tachyon model anthropic:claude-sonnet-5

To override for a single session without touching the default:

bash

tachyon --model openai:gpt-5

A practical split: a strong model for the main loop, and cheaper models for subagents doing mechanical work. Subagent presets already pick sensible tiers — see Subagents.

OpenAI-compatible endpoints

Anything speaking the OpenAI API — a corporate gateway, a router like OpenRouter, a self-hosted inference server — works by supplying a base URL at login time:

bash

tachyon login mygateway --base-url https://llm.internal.example.com/v1

This is also the route for providers that are not in the models.dev catalogue, or for pinning traffic to a region-specific endpoint.

Local models

Ollama is supported as a provider, so a model running on your own hardware needs no key and no network egress at all. This pairs naturally with the default network-denied sandbox: the agent can work on a codebase with no outbound connectivity whatsoever.