Providers

Coderrr reaches five providers over plain HTTP. No provider SDK is installed, and your key goes straight from your machine to the provider — nothing is proxied through a server.

At a glance

Provider	Key needed	Default model
Ollama	no — default provider	gemma4:31b-cloud
Anthropic	yes	claude-sonnet-4-20250514
OpenAI	yes	gpt-4o
Google	yes	gemini-2.0-flash
OpenRouter	yes	qwen/qwen-2.5-coder-32b-instruct

Tool use is the thing that matters

Coderrr is a real agent loop: the model reads files, calls tools, observes results, and iterates. A model that is merely good at writing code but weak at multi-turn tool use will struggle regardless of its benchmark scores.

Ollama — the default

Ollama

Local models or Ollama-hosted cloud models. No API key is required for local models; Ollama-hosted models use your Ollama account.

Install Ollama: ollama.com/download

Model	Notes
gemma4:31b-cloud	The out-of-the-box default. Tool-capable and usable on a free Ollama account.
kimi-k3:cloud	Stronger, 1M context — but returns 403 without a paid Ollama subscription, which is why it cannot be the default. Switch to it with coderrr config if you have one.
qwen2.5-coder:14b	Fully local, no account at all. Noticeably weaker at the multi-turn tool use the agent loop depends on.
qwen2.5-coder:7b	Fully local and smaller still. Same caveat, more so.

# fully local
ollama pull qwen2.5-coder:14b

coderrr config
# Select "Ollama", then a model
# Endpoint defaults to http://localhost:11434/v1

The endpoint prompt lets you point at a remote Ollama host. It rejects anything that is not a URL — a stray y stored there would make every later request fail in a way that reads like a network fault.

Hosted providers

Anthropic

Claude models. Strong at multi-turn tool use, which is what this agent loop leans on hardest.

Get a key: console.anthropic.com · Env var: ANTHROPIC_API_KEY

Model	Notes
claude-sonnet-4-20250514	Default. The balanced choice.
claude-opus-4-20250514	More capable, more expensive.

OpenAI

GPT models.

Get a key: platform.openai.com/api-keys · Env var: OPENAI_API_KEY

Model	Notes
gpt-4o	Default.
gpt-4o-mini	Cheaper. A good choice for [verify].model.

Google Gemini

Gemini models.

Get a key: aistudio.google.com/apikey · Env var: GOOGLE_API_KEY

Model	Notes
gemini-2.0-flash	Default. Fast and inexpensive.
gemini-1.5-pro	Larger context.

OpenRouter

Many providers behind one key, including free tiers. Useful for trying models without signing up for each vendor.

Get a key: openrouter.ai/keys · Env var: OPENROUTER_API_KEY

Model	Notes
qwen/qwen-2.5-coder-32b-instruct	Default.
anthropic/claude-3.5-sonnet	Claude through OpenRouter.
deepseek/deepseek-chat	Capable and cheap.

Any model ID OpenRouter accepts works — the suggestions are just starting points. Requests carry a HTTP-Referer and X-Title header for attribution on OpenRouter's dashboards.

Switching provider or model

coderrr config                  # change provider, model, or key
coderrr config show             # confirm what is active
coderrr run "..." -m gpt-4o     # override the model for one run
coderrr                         # then /model gpt-4o inside the session

A /model change inside a session applies to that session only and is not saved.

Custom endpoints

Ollama, OpenAI, and OpenRouter all go through one OpenAI-compatible adapter; Anthropic and Google need bespoke wire translation. If you run an OpenAI-compatible gateway, configure the Ollama provider and point its endpoint at your gateway.

[provider]
name     = "ollama"
model    = "your-model-id"
endpoint = "https://your-gateway.example.com/v1"

Removed in v2

Azure AI / GitHub Models is no longer a provider, and the GITHUB_TOKEN setting is gone with it. v1 reached those models through a hosted backend that v2 does not have. Use OpenRouter if you want many models behind one key.