Configuration

Vibe Action uses a single YAML config file at ~/.vibe-action/config.yaml. It is created automatically on first run.

Action

Runtime settings for all flows.

action:
  system: |
    You are Vibe Action — a CLI tool, not a chatbot.
    Work fast. Don't think too much. Just do the task.
    Output ONLY the requested result.
  retries: 2

Field	Type	Description
`system`	string	Global system prompt for all LLM requests
`retries`	integer	Number of retries for failed LLM steps (0 = off)

Cluster

Define one or more LLM providers. The engine sends prompts to all nodes in parallel. Use role to assign models to specific complexity levels.

cluster:
  - provider: ollama
    host: http://localhost:11434
    model: qwen2.5-coder:14b-instruct
    role: medium
    timeout_secs: 60
    temperature: 0.1
    seed: 42
    num_ctx: 4096
    num_predict: 2048
    parallel: 1

Cluster Node Fields

Field	Type	Description
`provider`	string	`ollama`, `deepseek`, `qwen`, `kimi`, `zhipu`
`host`	string	API endpoint URL
`model`	string	Model name
`role`	string	`tiny`, `small`, `medium`, `large`, `vision`
`timeout_secs`	integer	Request timeout in seconds
`temperature`	float	0.0-2.0, lower = more deterministic
`seed`	integer	Random seed for reproducibility
`num_ctx`	integer	Context window size in tokens
`num_predict`	integer	Max tokens to generate
`api_key`	string	API key for cloud providers (optional)
`parallel`	integer	Concurrent connections (default: 1)

Multi-Node Cluster with Roles

cluster:
  - provider: ollama
    host: http://localhost:11434
    model: qwen2.5-coder:3b-instruct
    role: tiny
    timeout_secs: 30
    temperature: 0.0
    seed: 42
    num_ctx: 4096
    num_predict: 512
    parallel: 2

  - provider: zhipu
    host: https://open.bigmodel.cn/api/paas/v4
    model: glm-4-flash
    role: small
    timeout_secs: 60
    temperature: 0.1
    seed: 42
    num_ctx: 8192
    num_predict: 2048
    api_key: sk-...
    parallel: 2

  - provider: ollama
    host: http://192.168.1.10:11434
    model: qwen2.5-coder:14b-instruct
    role: medium
    timeout_secs: 60
    temperature: 0.1
    seed: 42
    num_ctx: 8192
    num_predict: 2048
    parallel: 1

  - provider: deepseek
    host: https://api.deepseek.com/v1
    model: deepseek-v4-flash
    role: large
    timeout_secs: 120
    temperature: 0.1
    seed: 42
    num_ctx: 16384
    num_predict: 8192
    api_key: sk-...
    parallel: 2

  - provider: kimi
    host: https://api.moonshot.cn/v1
    model: moonshot-v1-8k
    role: large
    timeout_secs: 120
    temperature: 0.1
    seed: 42
    num_ctx: 8192
    num_predict: 2048
    api_key: sk-...
    parallel: 1

  - provider: ollama
    host: http://localhost:11434
    model: qwen2.5vl:7b
    role: vision
    timeout_secs: 120
    temperature: 0.1
    seed: 42
    num_ctx: 4096
    num_predict: 2048
    parallel: 1

Nodes with role: tiny are used for run: tiny, role: small for run: small, role: medium for run: medium, role: large for run: large, role: vision for run: vision. Nodes without a role respond to all requests.

Actions Directory

Actions are stored in ~/.vibe-action/actions/. The directory is created on first run with default actions. Override with VIBE_ACTION_PATH.

~/.vibe-action/
├── config.yaml
└── actions/
    ├── comment.yaml
    ├── commit.yaml
    ├── describe.yaml
    ├── explain.yaml
    ├── ...

Add your own .yaml files here — they will be loaded automatically.