$ docker run --rm -v ./my-config:/home/axe/.config/axe -e ANTHROPIC_API_KEY axe run my-agent

Updated6 days ago

LanguageGo

Setup difficultyEasy

Stars: 907

What is Axe?

Axe is a Go command-line tool for defining and running small, single-purpose LLM agents, in the Unix spirit: each agent does one job, is described in a version-controllable TOML file, and is invoked from the terminal, a pipe, a git hook, or cron. Agents carry their own system prompt, model choice, skill files, working directory, persistent timestamped memory, and can delegate work to sub-agents with depth limits and parallel execution. Built-in tools let an agent read, write, and edit files sandboxed to its working directory, run shell commands, fetch URLs, and search the web, with an hostname allowlist and SSRF protections plus optional per-run token budgets; external tools can be added via MCP servers. It talks to Anthropic, OpenAI, Ollama, OpenCode, AWS Bedrock, and Atlas Cloud, using only four direct Go dependencies. You install it as a pre-built binary for Linux, macOS, and Windows, via go install, or in a hardened non-root Docker container with an optional Ollama sidecar through the included compose file. It is an executor, not a scheduler: there is no daemon, GUI, or built-in workflow orchestration.

Should you self-host Axe?

Axe scores 51/100 on selfhost.directory (how scoring works) and ranks #94 of 209 in AI Agents. It has 907 GitHub stars.

It shipped 10 stable releases in the past year. The latest, v1.11.0, came out on 22 October 2026. New versions land roughly every 5 days.

Setup is easy: most people get it running in one sitting. It is Apache-2.0-licensed and written mostly in Go.

Its closest alternative on selfhost.directory is Orca (91/100, 108 446 stars). It scores higher overall, so compare the two before you pick. See Axe vs Orca side by side.

Features

Built-in agent tools
Configurable retry
Dry Run Mode
JSON Output
Local agent directory discovery
MCP tool support
Memory garbage collection
Multi-Provider LLM Support
Output hostname allowlist
Persistent memory
Reusable skill system
Stdin piping
Sub-agent delegation
Token budget per run
TOML agent configuration

How to configure Axe

--agents-dir(auto-discover)Override agent search directory
Override agent search directory
ANTHROPIC_API_KEYAPI authentication
API authentication
AXE_ANTHROPIC_BASE_URLOverride Anthropic API endpoint
Override Anthropic API endpoint
AXE_OLLAMA_BASE_URLOllama endpoint (default in compose: `http://ollama:11434`)
Ollama endpoint (default in compose: `http://ollama:11434`)
AXE_OPENAI_BASE_URLOverride OpenAI API endpoint
Override OpenAI API endpoint
AXE_OPENCODE_BASE_URLOverride OpenCode API endpoint
Override OpenCode API endpoint
AXE_WEB_SEARCH_BASE_URLOverride web search endpoint
Override web search endpoint
backoff"exponential"Strategy: `"exponential"` (with jitter), `"linear"`, or `"fixed"`
Strategy: `"exponential"` (with jitter), `"linear"`, or `"fixed"`
--dry-runfalseShow resolved context without calling the LLM
Show resolved context without calling the LLM
headersHTTP headers; values support `${ENV_VAR}` interpolation
HTTP headers; values support `${ENV_VAR}` interpolation
initial_delay_ms500Base delay in milliseconds before the first retry
Base delay in milliseconds before the first retry
--jsonfalseWrap output in a JSON envelope with metadata
Wrap output in a JSON envelope with metadata
max_delay_ms30000Maximum delay cap in milliseconds
Maximum delay cap in milliseconds
max_retries0Number of retry attempts after the initial request. 0 disables retry.
Number of retry attempts after the initial request. 0 disables retry.
--max-tokens0 (unlimited)Cap cumulative token usage for the run (exit code 4 if exceeded)
Cap cumulative token usage for the run (exit code 4 if exceeded)
--modelfrom TOMLOverride the model (e.g. `anthropic/claude-sonnet-4-20250514`)
Override the model (e.g. `anthropic/claude-sonnet-4-20250514`)
nameHuman-readable identifier for the server
Human-readable identifier for the server
OPENAI_API_KEYAPI authentication
API authentication
-p(none)Inline prompt used as the user message; takes precedence over stdin
Inline prompt used as the user message; takes precedence over stdin
--skillfrom TOMLOverride the skill file path
Override the skill file path
TAVILY_API_KEYTavily web search API key
Tavily web search API key
--timeout120Request timeout
Request timeout
transport`"sse"` or `"streamable-http"`
`"sse"` or `"streamable-http"`
urlMCP server endpoint URL
MCP server endpoint URL
--verbosefalsePrint debug info (model, timing, tokens, retries) to stderr
Print debug info (model, timing, tokens, retries) to stderr
--workdirfrom TOML or cwdOverride the working directory
Override the working directory