Setup and Local Models

Use this page after installation to configure API keys, provider defaults, and OpenAI-compatible profile records.

Setup wizard

The setup wizard is the recommended path for most users.

fastfold setup

export ANTHROPIC_API_KEY="sk-ant-..."
export OPENAI_API_KEY="sk-..."
export FASTFOLD_API_KEY="sk-..."
export BOLTZ_API_KEY="sk_bc_..."

fastfold setup --provider anthropic --api-key sk-ant-... --fastfold-api-key sk-...
fastfold setup --provider openai --openai-api-key sk-... --fastfold-api-key sk-...
fastfold setup --provider anthropic --api-key sk-ant-... --fastfold-api-key sk-... --boltz-api-key sk_bc_...
fastfold setup --provider openai_compatible --profile-label "oMLX Local" --profile-template omlx --profile-endpoint http://localhost:8000/v1 --profile-key sk-...
fastfold setup --provider openai_compatible --profile-label "DS4 Local" --profile-template ds4 --profile-endpoint http://localhost:8000/v1 --profile-key dsv4-local
fastfold setup --provider openai_compatible --profile-label "llama.cpp Local" --profile-template llama_cpp --profile-endpoint http://localhost:8080/v1
fastfold setup --provider openai_compatible --profile-label "LM Studio Local" --profile-template lm_studio --profile-endpoint http://localhost:1234/v1

# Skills and datasets non-interactively
fastfold setup --provider anthropic --api-key sk-ant-... --skills "fastfold-ai/skills@skills/fold" --datasets depmap,msigdb
fastfold setup --provider anthropic --api-key sk-ant-... --skip-skills --skip-datasets

Optional setup steps

After keys, the wizard offers two optional steps (skip with flags for CI):

• Skills — multi-select from the live catalog. Non-interactive: --skills "<sources>" or --skip-skills.
• Datasets — multi-select of auto-downloadable datasets, all preselected. Non-interactive: --datasets depmap,msigdb, --datasets all, or --skip-datasets. See Data and reports.
• Boltz stack (when BOLTZ_API_KEY is configured) — offers Boltz skill install plus boltz-api CLI install, then validates in doctor output.

Provider configuration

Use these commands when you want explicit provider/model settings in config.

fastfold config set llm.provider anthropic
fastfold config set llm.model claude-sonnet-4-5-20250929
fastfold config set llm.anthropic_api_key sk-ant-...

fastfold config set llm.provider openai
fastfold config set llm.model gpt-5.5
fastfold config set llm.openai_api_key sk-...

fastfold setup --provider openai_compatible \
  --profile-label "Ollama Local" \
  --profile-template ollama \
  --profile-endpoint http://localhost:11434/v1 \
  --profile-key ollama

Provider integrations (/keys)

Use /keys (interactive) or fastfold keys (shell) to inspect provider status and config key mappings.

Boltz quick setup inside interactive mode:

/keys set-boltz

Configured tool keys are mirrored into the session environment on startup, so shell/skill subprocesses can read BOLTZ_API_KEY and FASTFOLD_API_KEY without extra export commands.

Local and OpenAI-compatible models

Use this for Ollama, Unsloth, oMLX, DS4, llama.cpp, LM Studio, or another OpenAI-compatible endpoint. Profiles are the system of record for compatible providers.

Start with the wizard:

fastfold setup --provider openai_compatible

The wizard asks for backend type, base URL, API key, and available model. It now also shows a summary (profile/template/endpoint) before model selection.

Typical defaults:

• Ollama: http://localhost:11434/v1
• Unsloth: http://localhost:8888/v1
• oMLX: http://localhost:8000/v1
• DS4: http://localhost:8000/v1
• llama.cpp: http://localhost:8080/v1
• LM Studio: http://localhost:1234/v1

Template values:

• ollama
• unsloth
• omlx
• ds4
• llama_cpp
• lm_studio
• other

Useful profile flags for scripted setup:

fastfold setup --provider openai_compatible \
  --profile-label "Custom Gateway" \
  --profile-template other \
  --profile-endpoint https://gateway.example.com/v1 \
  --profile-key sk-... \
  --profile-default-model gpt-oss-120b \
  --set-default-profile

Interactive workflow:

• /model selects from configured models/providers (selection only).
• /model-manager adds/edits/deletes compatible profiles and shows endpoint health + discovered models.
• /keys set-compatible <profile_id> updates a specific compatible profile key.
• /keys set-boltz updates BOLTZ_API_KEY and offers Boltz skill + CLI installation.

When a local compatible backend is active (for example Ollama) but a cloud-only model id remains selected (gpt-*), the CLI auto-heals to the active profile default model to prevent 404 model-not-found loops.

If /model-manager shows no models for Unsloth, auth is usually valid but no model is loaded in Unsloth Studio yet.

Inference engine install references

Before creating compatible profiles in Fastfold, install/run one of these local inference engines:

• DS4 (DeepSeek v4 Flash/PRO): antirez/ds4
• llama.cpp: ggml-org/llama.cpp
• LM Studio (OpenAI-compatible server): OpenAI Compatibility Endpoints
• Ollama: ollama/ollama
• oMLX: jundot/omlx
• Unsloth: unslothai/unsloth

Related pages

• Install and upgrade overview: Agent CLI Overview
• First query and health checks: Usage and quick start
• Interactive commands: Interactive commands