What problem does it solve? Running NanoClaw agents against the Anthropic API incurs per-token costs and requires network access. This Skill reconfigures an existing agent group to use a locally running Ollama model instead, cutting API spend and enabling experimentation with open-weight models without changing provider code. ## Core Features & Use Cases - Env-based rerouting: Sets ANTHROPIC_BASE_URL, ANTHROPIC_API_KEY, and NO_PROXY in the group's container.json so the Anthropic SDK talks to Ollama's native v1/messages endpoint. - Source wiring when missing: Adds env and blockedHosts fields to ContainerConfig and container-runner.ts if the codebase lacks them, including Docker --add-host blocking of api.anthropic.com. - Model configuration and verification: Writes the model name into the group's shared Claude settings.json, rebuilds and restarts the service, and verifies via Ollama and Docker inspection commands. - Use Case: You want your Telegram-facing agent to run on a local gemma4 model with zero Anthropic spend; the Skill configures the group, blocks api.anthropic.com as a safeguard, and confirms the model is active. ## Quick Start Ask the assistant to switch a specific NanoClaw agent group to a local Ollama model such as gemma4 and block the Anthropic API.
