LLM_PROVIDERS

LLM Providers

How the Dev Companion selects and uses AI models.

---

Provider priority

The runner automatically selects providers in this order:

Priority	Provider	Type	Cost	Requirements
1	OpenCode (big-pickle)	Local	Free	opencode installed
2	Ollama	Local	Free	ollama installed + models
3	Anthropic (Claude)	Cloud	Paid	ANTHROPIC_API_KEY
4	OpenAI	Cloud	Paid	OPENAI_API_KEY

---

Zero-config experience

For most developers:

1. Install OpenCode → Done
2. Run jobs → Works immediately with big-pickle

No API keys required. No configuration files.

---

Advanced: Ollama (Local GPU)

brew install ollama           # macOS
curl -fsSL https://ollama.com/install.sh | sh  # Linux

ollama pull llama3.2          # pull a model
# The runner auto-detects and uses it

---

Advanced: Cloud providers

export ANTHROPIC_API_KEY="sk-ant-..."   # Claude
export OPENAI_API_KEY="sk-..."          # OpenAI

Cloud providers are used only if no free local provider is available.

---

Debugging

DOTS_WORKSTATION_LLM_DEBUG=1 dots-devcompanion run-once   # debug logging
opencode --version                           # check OpenCode
ollama list                                  # check Ollama models

---

Fallback behavior

Scenario	Behavior
No providers available	Generates skeleton plan
Provider times out	Writes error in artifacts
Provider returns invalid response	Falls back to skeleton

---

Canonical doc: docs/DEV_COMPANION_LLM.md