Home / Providers

Five providers, switched at runtime

tokenworm speaks to five LLM providers through one vtable-based abstraction. Switch with --provider on the CLI or switch_provider() in the SDK — no code change, no rebuild. Every provider shares the same agent loop, tool dispatch, and .tworm sessions.

OpenCode Zen

Default

The default provider — curated free models for coding agents, so you can try tokenworm without wiring up a paid key first.

OPENCODE_ZEN_API_KEY

Anthropic

Hosted

Anthropic Claude models behind the same agent loop and tool dispatch. One of the five providers tokenworm speaks to.

ANTHROPIC_API_KEY

OpenAI

Hosted

OpenAI models for broad model selection, driven through the same provider abstraction.

OPENAI_API_KEY

Ollama

Local

Run entirely on your own machine with no API key — private, offline-capable, and driven by the same core.

install ollama

MiniMax

Hosted

An OpenAI-compatible alternative provider, selectable at runtime like the rest.

MINIMAX_API_KEY

Switch at runtime

~ tokenworm

# CLI: pick a provider per invocation
tokenworm --provider ollama "summarize this repo"

# SDK: switch without restarting the agent
agent.switch_provider("anthropic")

Provider selection is a runtime concern; the tools, sandbox, and session format stay the same across all five.