BUILD / COMMAND-LINE INTERFACES

Rust CLI.
Command-line reference.

Run agentic in your terminal to start the gateway and launch Codex or Claude Code. Available through Cargo and the PyPI package.

Source-generated reference
Read as MarkdownView CLI source

The $ marks a terminal prompt; leave it out when typing commands.

Generated from the CLI source in this checkout (package version 0.7.0). Use agentic --help or agentic <command> --help for your installed version.

The Rust command-line interface is agentic .... Use it from a terminal to start the gateway, launch Codex or Claude Code, attach a harness to an existing gateway, and validate your configuration.

Both cargo install agentic-server and pip install agentic-api install agentic. The PyPI package also includes the separate Python launcher, invoked as agentic-api ... or python -m agentic_api .... Its commands, flags, and defaults differ from the Rust CLI.

See Python installation and workflows for package installation.

agentic

Agentic API — local agent gateway for Claude Code and Codex

Usage: agentic <COMMAND>

Commands:
  run       Start Agentic API and launch a coding harness
  harness   Launch a coding harness against an already-running Agentic API gateway
  serve     Start Agentic API without launching a harness
  validate  Validate the local Agentic API session prerequisites
  help      Print this message or the help of the given subcommand(s)

Options:
  -h, --help
          Print help

  -V, --version
          Print version

agentic run

Start Agentic API and launch a coding harness

Usage: agentic run <COMMAND>

Commands:
  codex   Launch Codex with an isolated provider configuration
  claude  Launch Claude Code with an isolated gateway environment
  help    Print this message or the help of the given subcommand(s)

Options:
  -h, --help
          Print help

agentic run codex

Launch Codex with an isolated provider configuration

Usage: agentic run codex [OPTIONS] [-- <HARNESS_ARGS>...]

Arguments:
  [HARNESS_ARGS]...
          Arguments forwarded to the selected harness after `--`

Options:
      --upstream <UPSTREAM>
          Connect to an already-running OpenAI-compatible upstream (`http://` or `https://` base URL)

      --model <MODEL>
          Model to start with vLLM, or the model name to use with `--upstream`. When omitted alongside `--upstream`, the first model served by the upstream is used

      --llm-port <LLM_PORT>
          vLLM port when starting a model

          [default: 8000]

      --gateway-host <GATEWAY_HOST>
          Gateway bind host

          [env: GATEWAY_HOST]
          [default: 127.0.0.1]

      --gateway-port <GATEWAY_PORT>
          Gateway bind port

          [env: GATEWAY_PORT]
          [default: 3000]

      --database-url <DATABASE_URL>
          `SQLite` or `PostgreSQL` storage URL

          [env: DATABASE_URL]
          [default: sqlite://./agentic_api.db]

      --max-request-body-size-bytes <MAX_REQUEST_BODY_SIZE_BYTES>
          Maximum serialized request size in bytes accepted by the gateway. Left to the gateway's own configuration chain when omitted

      --api-key <API_KEY>
          API key forwarded to the gateway and harness when configured

          [env: OPENAI_API_KEY]

      --skip-llm-ready-check
          Skip the upstream readiness probe

      --llm-ready-timeout-s <LLM_READY_TIMEOUT_S>
          Upstream readiness timeout in seconds

          [default: 600]

      --llm-ready-interval-s <LLM_READY_INTERVAL_S>
          Upstream readiness poll interval in seconds

          [default: 2]

      --quiet
          Suppress lifecycle output

      --yolo
          Skip harness permission prompts and sandbox restrictions

      --no-color
          Disable ANSI color output

  -h, --help
          Print help

agentic run claude

Launch Claude Code with an isolated gateway environment

Usage: agentic run claude [OPTIONS] [-- <HARNESS_ARGS>...]

Arguments:
  [HARNESS_ARGS]...
          Arguments forwarded to the selected harness after `--`

Options:
      --upstream <UPSTREAM>
          Connect to an already-running OpenAI-compatible upstream (`http://` or `https://` base URL)

      --model <MODEL>
          Model to start with vLLM, or the model name to use with `--upstream`. When omitted alongside `--upstream`, the first model served by the upstream is used

      --llm-port <LLM_PORT>
          vLLM port when starting a model

          [default: 8000]

      --gateway-host <GATEWAY_HOST>
          Gateway bind host

          [env: GATEWAY_HOST]
          [default: 127.0.0.1]

      --gateway-port <GATEWAY_PORT>
          Gateway bind port

          [env: GATEWAY_PORT]
          [default: 3000]

      --database-url <DATABASE_URL>
          `SQLite` or `PostgreSQL` storage URL

          [env: DATABASE_URL]
          [default: sqlite://./agentic_api.db]

      --max-request-body-size-bytes <MAX_REQUEST_BODY_SIZE_BYTES>
          Maximum serialized request size in bytes accepted by the gateway. Left to the gateway's own configuration chain when omitted

      --api-key <API_KEY>
          API key forwarded to the gateway and harness when configured

          [env: OPENAI_API_KEY]

      --skip-llm-ready-check
          Skip the upstream readiness probe

      --llm-ready-timeout-s <LLM_READY_TIMEOUT_S>
          Upstream readiness timeout in seconds

          [default: 600]

      --llm-ready-interval-s <LLM_READY_INTERVAL_S>
          Upstream readiness poll interval in seconds

          [default: 2]

      --quiet
          Suppress lifecycle output

      --yolo
          Skip harness permission prompts and sandbox restrictions

      --no-color
          Disable ANSI color output

  -h, --help
          Print help

agentic harness

Launch a coding harness against an already-running Agentic API gateway

Usage: agentic harness <COMMAND>

Commands:
  codex   Launch Codex with an isolated provider configuration
  claude  Launch Claude Code with isolated provider and model configuration
  help    Print this message or the help of the given subcommand(s)

Options:
  -h, --help
          Print help

agentic harness codex

Launch Codex with an isolated provider configuration

Usage: agentic harness codex [OPTIONS] --gateway-url <GATEWAY_URL> --model <MODEL> [-- <HARNESS_ARGS>...]

Arguments:
  [HARNESS_ARGS]...
          Arguments forwarded to the selected harness after `--`

Options:
      --gateway-url <GATEWAY_URL>
          URL of an already-running Agentic API gateway

      --model <MODEL>
          Model ID served by the gateway

      --api-key <API_KEY>
          API key forwarded to the gateway and harness when configured

          [env: AGENTIC_GATEWAY_API_KEY]

      --quiet
          Suppress lifecycle output

      --yolo
          Skip harness permission prompts and sandbox restrictions

      --no-color
          Disable ANSI color output

  -h, --help
          Print help

agentic harness claude

Launch Claude Code with isolated provider and model configuration

Usage: agentic harness claude [OPTIONS] --gateway-url <GATEWAY_URL> --model <MODEL> [-- <HARNESS_ARGS>...]

Arguments:
  [HARNESS_ARGS]...
          Arguments forwarded to the selected harness after `--`

Options:
      --gateway-url <GATEWAY_URL>
          URL of an already-running Agentic API gateway

      --model <MODEL>
          Model ID served by the gateway

      --api-key <API_KEY>
          API key forwarded to the gateway and harness when configured

          [env: AGENTIC_GATEWAY_API_KEY]

      --quiet
          Suppress lifecycle output

      --yolo
          Skip harness permission prompts and sandbox restrictions

      --no-color
          Disable ANSI color output

  -h, --help
          Print help

agentic serve

Start Agentic API without launching a harness

Usage: agentic serve [OPTIONS]

Options:
      --upstream <UPSTREAM>
          Connect to an already-running OpenAI-compatible upstream (`http://` or `https://` base URL)

      --model <MODEL>
          Model to start with vLLM, or the model name to use with `--upstream`. When omitted alongside `--upstream`, the first model served by the upstream is used

      --llm-port <LLM_PORT>
          vLLM port when starting a model

          [default: 8000]

      --gateway-host <GATEWAY_HOST>
          Gateway bind host

          [env: GATEWAY_HOST]
          [default: 127.0.0.1]

      --gateway-port <GATEWAY_PORT>
          Gateway bind port

          [env: GATEWAY_PORT]
          [default: 3000]

      --database-url <DATABASE_URL>
          `SQLite` or `PostgreSQL` storage URL

          [env: DATABASE_URL]
          [default: sqlite://./agentic_api.db]

      --max-request-body-size-bytes <MAX_REQUEST_BODY_SIZE_BYTES>
          Maximum serialized request size in bytes accepted by the gateway. Left to the gateway's own configuration chain when omitted

      --api-key <API_KEY>
          API key forwarded to the gateway and harness when configured

          [env: OPENAI_API_KEY]

      --skip-llm-ready-check
          Skip the upstream readiness probe

      --llm-ready-timeout-s <LLM_READY_TIMEOUT_S>
          Upstream readiness timeout in seconds

          [default: 600]

      --llm-ready-interval-s <LLM_READY_INTERVAL_S>
          Upstream readiness poll interval in seconds

          [default: 2]

      --quiet
          Suppress lifecycle output

      --yolo
          Skip harness permission prompts and sandbox restrictions

      --no-color
          Disable ANSI color output

  -h, --help
          Print help

agentic validate

Validate the local Agentic API session prerequisites

Usage: agentic validate [OPTIONS]

Options:
      --upstream <UPSTREAM>
          Connect to an already-running OpenAI-compatible upstream (`http://` or `https://` base URL)

      --model <MODEL>
          Model to start with vLLM, or the model name to use with `--upstream`. When omitted alongside `--upstream`, the first model served by the upstream is used

      --llm-port <LLM_PORT>
          vLLM port when starting a model

          [default: 8000]

      --gateway-host <GATEWAY_HOST>
          Gateway bind host

          [env: GATEWAY_HOST]
          [default: 127.0.0.1]

      --gateway-port <GATEWAY_PORT>
          Gateway bind port

          [env: GATEWAY_PORT]
          [default: 3000]

      --database-url <DATABASE_URL>
          `SQLite` or `PostgreSQL` storage URL

          [env: DATABASE_URL]
          [default: sqlite://./agentic_api.db]

      --max-request-body-size-bytes <MAX_REQUEST_BODY_SIZE_BYTES>
          Maximum serialized request size in bytes accepted by the gateway. Left to the gateway's own configuration chain when omitted

      --api-key <API_KEY>
          API key forwarded to the gateway and harness when configured

          [env: OPENAI_API_KEY]

      --skip-llm-ready-check
          Skip the upstream readiness probe

      --llm-ready-timeout-s <LLM_READY_TIMEOUT_S>
          Upstream readiness timeout in seconds

          [default: 600]

      --llm-ready-interval-s <LLM_READY_INTERVAL_S>
          Upstream readiness poll interval in seconds

          [default: 2]

      --quiet
          Suppress lifecycle output

      --yolo
          Skip harness permission prompts and sandbox restrictions

      --no-color
          Disable ANSI color output

      --harness <HARNESS>
          Also verify a harness binary without launching it

          [possible values: codex, claude]

  -h, --help
          Print help