AutoDevOS CLI Docs

System Requirements

1. Run the Setup Wizard

On first use, run the interactive auth wizard to configure your LLM provider and API key:

ados auth setup

The wizard walks through selecting a provider, entering your API key, and setting a default model.

2. Open a Project

Navigate to any codebase and start an interactive session:

cd my-project
ados

You can also pass a directory directly, or run a single prompt without entering interactive mode:

ados ~/projects/my-app

ados "refactor the authentication module"

3. Describe Your Task

Type any natural language task at the prompt. The agent reads your project, selects tools, and executes autonomously:

> add input validation to the /register endpoint and write tests for it

AGENTS.md

Drop an AGENTS.md file at your repo root or in any subdirectory. The agent reads all matching files in the current path hierarchy and uses them as project-level instructions — coding conventions, testing notes, architecture decisions, and more.

All Config Fields

Switching Providers

ados config provider anthropic
ados config model claude-opus-4-5

Local Providers (Ollama, LM Studio, vLLM)

Local providers need no API key. If your server runs on a non-default port, set BASE_URL:

ados config provider ollama
BASE_URL=http://localhost:11434 ados

Custom OpenAI-Compatible Endpoint

[model]
provider = "custom"
name = "my-model"

[model]
base_url = "https://my-llm-proxy.example.com/v1"

Main Commands

Config Subcommands

ados config              # show active configuration
ados config provider     # list or set the LLM provider
ados config model        # set the model name
ados config url          # set a custom base URL
ados config providers    # list all available providers

Auth Subcommands

ados auth                         # show current auth status
ados auth setup                   # run the interactive setup wizard
ados auth set-key <provider>      # store an API key for a provider
ados auth remove-key <provider>   # remove a stored key
ados auth generate-token          # generate a shareable team setup token
ados auth setup-token <token>     # apply a team setup token

Restricting Tools per Project

Limit the tools available in a project by setting allowed_tools in .autodevos/config.toml:

allowed_tools = ["read_file", "edit", "grep", "shell"]

Stdio Transport

Launch an MCP server as a child process over stdin/stdout:

[mcp_servers.filesystem]
type = "stdio"
command = "npx"
args = ["-y", "@modelcontextprotocol/server-filesystem", "/path/to/dir"]

HTTP/SSE Transport

Connect to a running MCP server over HTTP Server-Sent Events:

[mcp_servers.my-server]
type = "sse"
url = "http://localhost:3100/sse"

Approval Policies

Control when the agent asks for your confirmation before executing tool calls:

approval = "auto-edit"

Blocked Shell Commands

The shell tool hard-blocks a set of destructive patterns regardless of approval policy:

• →rm -rf / and similar root-level deletions
• →Fork bombs: :(){ :|:&};:
• →Filesystem format commands: mkfs.*
• →Other hard-coded dangerous patterns

Secret Variable Filtering

Environment variables matching *KEY*, *TOKEN*, and *SECRET* are stripped from the shell environment visible to the agent. Override or extend this list via the shell_environmentconfig section.

Path Validation

File read and write tools validate paths to prevent directory traversal outside the working directory.

Context Compaction

When a conversation approaches the configured context_window, AutoDevOS automatically summarises older messages into a compact representation so the session continues without losing task context. Increase the window for models that support it:

[model]
context_window = 200000

Persistent Memory Tool

The memory tool gives the agent a persistent key-value store that survives across sessions, stored at ~/.local/share/autodevos/user_memory.json. Use it to remember preferences, project notes, or any context the agent should recall in future sessions.