RACT

Rooted by Dr. Lucas Root, Ph.D.

RootAct Provider Setup Guide

RootAct routes each plan step to a provider adapter. You can use local models, remote APIs, or multiple providers side by side.

Supported adapters

Adapter	Use case	Configuration keys
local_http	Local servers such as llama-server, ollama, lmstudio	url, model
openai	OpenAI-compatible cloud APIs	url, api_key, model

Local model via llama-server

Start the server:

llama-server -m /path/to/model.gguf --port 8011 --host 127.0.0.1 -c 32768

Configure RootAct:

manager_provider: local

providers:
  local:
    adapter: local_http
    url: http://127.0.0.1:8011/v1
    model: my-local-model

The model value is passed to the server but may be ignored by local servers.

Ollama

providers:
  ollama:
    adapter: local_http
    url: http://127.0.0.1:11434/v1
    model: llama3.1

Local model behind an inference proxy

If you run a local inference proxy that routes to a model on port 11434:

providers:
  local_proxy:
    adapter: local_http
    url: http://127.0.0.1:11434/v1
    model: my-local-model

OpenAI API

providers:
  openai:
    adapter: openai
    url: https://api.openai.com/v1
    api_key: ${OPENAI_API_KEY}
    model: gpt-4o-mini

Cheap frontier providers

Z.ai

providers:
  zai:
    adapter: openai
    url: https://api.z.ai/v1
    api_key: ${ZAI_API_KEY}
    model: your-zai-model

Moonshot AI

providers:
  moonshot:
    adapter: openai
    url: https://api.moonshot.cn/v1
    api_key: ${MOONSHOT_API_KEY}
    model: moonshot-v1-8k

OpenRouter

providers:
  openrouter:
    adapter: openai
    url: https://openrouter.ai/api/v1
    api_key: ${OPENROUTER_API_KEY}
    model: openai/gpt-4o-mini

RootAct expands ${...} placeholders from the environment at runtime. Keep secrets out of the config file by using environment variables.

Multiple providers

You can define more than one provider. The manager uses manager_provider to choose which model plans the work. Each step can then request a provider by hint:

manager_provider: planner

providers:
  planner:
    adapter: openai
    url: https://api.openai.com/v1
    api_key: ${OPENAI_API_KEY}
    model: gpt-4o
  coder:
    adapter: openai
    url: https://api.openai.com/v1
    api_key: ${OPENAI_API_KEY}
    model: gpt-4o-mini

In a plan, a step can specify provider_hint: coder. RootAct’s router selects the best matching provider.

Retry and timeout behavior

Providers support configurable retries:

providers:
  openai:
    adapter: openai
    url: https://api.openai.com/v1
    api_key: ${OPENAI_API_KEY}
    model: gpt-4o-mini
    max_retries: 3
    retry_delay: 1.0
    retry_backoff: 2.0
    retry_max_delay: 30.0
    retry_on_429: true

• max_retries: how many times to retry a failed request.
• retry_delay: initial delay in seconds.
• retry_backoff: multiplier applied after each retry.
• retry_max_delay: cap on the delay between retries.
• retry_on_429: whether to retry HTTP 429 rate-limit responses.

Streaming

If the adapter advertises the streaming capability, RootAct can stream deltas as they arrive:

rootact "explain this code" --config rootact.yaml --stream

Streaming is optional; if the adapter does not support it, RootAct falls back to a non-streaming completion.

Capability-based routing

The router scores providers by how well they match a step’s hint. A provider named or tagged for a hint is preferred. If no specific match exists, the first configured provider is used.

Security notes

• Never commit API keys to version control.
• Use environment variables or a secrets manager.
• RootAct’s safety guardrails block eval(), exec(), subprocess...shell=True, and bare except: in generated content.

This site is open source. Improve this page.