Skip to content
Free and open source

Docs

AI assistants carry out coding tasks. This guide covers command-line tools (CLI), terminal menus (TUI), the terminal tool that keeps tasks running (tmux), and a standard connection between AI tools (MCP).

Install

You need macOS, Linux or WSL, Node 22.12 or newer, tmux 3.0 or newer, and at least one provider CLI that is installed and signed in.

$ pnpm add -g reevesagents$ reevesagents doctor$ reevesagents
reevesagents doctor checks all of that. reevesagents setup runs the same checks as a first-run wizard, and with --attach it also connects the MCP to every installed host.

Quick start

Start a team, check on it, step in, and stop it.

Start a team

shell
reevesagents spawn cc:lead codex:api kimi:docs \
  --name "feature x" --prompt "Build feature X. The lead splits the work."

Each assistant is written as provider[:nickname[:model]], and the first one leads. Use reevesagents add to add assistants to the latest run.

Watch and steer

shell
reevesagents runs
reevesagents peek api -n 40
reevesagents send api "run the tests"
reevesagents key api enter
reevesagents open api

send pastes text without submitting it, so follow it with key <agent> enter. open takes you to the assistant in tmux.

Stop

shell
reevesagents stop "feature x" --yes
stop, kill and the delete commands need --yes (or ALLOW_DESTRUCTIVE=1).

Terminal menu, browser and commands

The menu (TUI), browser interface and command-line tool (CLI) share task records under ~/.reeves. The terminal tool tmux keeps assistants running.

  • Type reevesagents with no arguments to open the TUI. It wraps itself in a tmux session called reeves, which REEVES_NO_TMUX_WRAPPER=1 turns off.
  • reevesagents web starts the Web UI on 127.0.0.1, port 8080 or the next free one. It refuses non-loopback Host headers and cross-origin writes, and Ctrl+C stops it without touching your assistants.
  • Everything is also a CLI command. Scripts should read --json, because the plain text output may change between versions.
  • Underneath sits tmux, with one session per run and one window per assistant. reevesagents open <id> jumps to the right window from wherever you are.

Agent Control (MCP)

Connect one coding tool to the others through a standard protocol (MCP). This is optional. Restart the tool after attaching it.

shell
reevesagents attach cc
reevesagents hosts
  • Six CLIs can host it: Claude Code, Codex, Kimi Code, Qwen Code, Hermes and OpenCode. reevesagents attach with no name attaches all the installed ones, and reevesagents detach <cli> removes it again.
  • The host gets plain tools to spawn assistants, paste text, press keys, read output and handle approvals. What the team does is decided by whoever prompts the host.
  • Spawned workers do not get the tools unless you attach their CLI too. MCP spawning enforces max_agents (100 by default) as the per-run assistant limit.

Commands

Every reevesagents subcommand. With no arguments it opens the TUI.

Runs

spawn [spec...]

Start a run. Each spec is provider[:nickname[:model]] and the first one leads. Flags: --name, --cwd, --prompt, --skip, --run, --auth-mode, --effort, --extra-args, --json.

add [spec...]

Add agents to the most recent active run. Flags: --run, --prompt, --skip, --auth-mode, --effort, --extra-args, --json. At least one spec is required.

runs

List active runs. --json.

open <id>

Jump to a run or agent window in tmux. Accepts an id, name, nickname or unique prefix.

peek <agent-id>

Print recent output from one agent. -n/--lines (default 20), --json.

stop <run-id>needs --yes

Stop one run and tear down its tmux session. -y/--yes.

kill <agent-id>needs --yes

Stop one agent. The run ends with its last live agent. -y/--yes.

Driving an assistant

send <agent-id> <text...>

Paste text at the agent prompt. Does not submit.

key <agent-id> <key>

Send one allowed key: enter, escape, backspace, tab, space, arrows, ctrl-c.

interrupt <agent-id>

Send ctrl-c.

Records and history

agents [run-id]

List agents across active runs or one specified run. --json.

providers

List providers with availability and known models. --models, --json.

tools

List the agent tools with install and sign-in status. --json.

delete <agent-id>needs --yes

Delete an ended agent record. -y/--yes.

delete-run <run-id>needs --yes

Delete an ended run. -y/--yes.

history

List archived run history. --json.

delete-history <id>needs --yes

Delete one history record. -y/--yes.

Approvals

approvals

List pending approvals. --json.

approve <approval-id> [note]

Resolve an approval as approved.

deny <approval-id> [note]

Resolve an approval as denied.

MCP hosts

hosts

List MCP-capable host CLIs and which are attached. --json.

attach [cli]

Attach the MCP to one host CLI, or all installed hosts, using its CLI or config file. --force rewrites an existing registration.

detach <cli>

Detach from one host CLI using its CLI or config file.

mcp

Start the Agent Control MCP server over stdio. The host CLI runs this itself.

Config and presets

config [key] [value]

Show, get, or set an editable config field. --json.

presets

List saved presets. --json.

save-preset <run-id> <name> [description...]

Capture a live run into a preset.

start-preset <name>

Start a run from a preset. --name, --cwd.

delete-preset <name>needs --yes

Delete a preset. -y/--yes.

Setup and tools

setup

First-run check of tmux, Node and installed CLIs. --attach connects every installed host. --json reports the onboarding state.

doctor

Run environment health checks. Non-zero exit on failure. --json.

skills [action]

Install, remove or check the ReevesAgents skill for skill-aware CLIs (install, remove, status; default status). --json.

reap

End agents whose window is gone or that passed max_lifetime_ms, and kill orphan tmux sessions. --json.

web

Start the loopback-only Web UI in the foreground. --port, --no-open.

MCP tools

The 31 tools an attached CLI can call, grouped by purpose.

Spawn and drive

spawnrequiredprovideroptionalrun_idmodeltasknicknamenameworking_dirpermissionsauth_modeeffortextra_args

Spawn a worker. Omit run_id to reuse this MCP session’s run, created on its first spawn; pass run_id to target another run.

killrequiredagent_id

Stop one agent.

stoprequiredrun_id

Stop a whole run and tear down its tmux session.

send_textrequiredagent_idtext

Paste text at an agent prompt. Does not submit.

send_keyrequiredagent_idkey

Send one allowed key: enter, escape, backspace, tab, space, arrows, ctrl-c.

interruptrequiredagent_id

Send ctrl-c to an agent.

readrequiredagent_idoptionallines

Read recent pane output, ANSI-stripped and redacted. Default: 20 lines.

Inspect

list

List runs and agents.

list_providers

Provider and model catalog for this machine.

list_tools

Tool catalog for this machine: helper CLIs an agent can run from its shell, with install and sign-in status.

list_history

Archived runs.

list_presets

Saved presets.

list_hosts

MCP-capable CLIs and which are attached.

get_config

Current global config.

doctor

Environment health checks.

Approvals

request_approvalrequiredagent_idactionsummaryoptionaldetailsrisk

Raise an approval request.

resolve_approvalrequiredapproval_iddecisionoptionalnote

Approve or deny a pending request.

check_approvalrequiredapproval_id

Read one approval.

list_approvalsoptionalrun_idstatus

List approvals, filterable.

Lifecycle

deleterequiredagent_id

Delete an ended agent record.

delete_runrequiredrun_id

Delete an ended run.

delete_historyrequiredid

Delete one history record.

openrequiredid

Switch the attached tmux client to a run or agent.

reap

Sweep dead-window and over-lifetime agents plus orphan tmux sessions now.

Config and presets

set_configoptionalpeek_interval_mspeek_linesmax_depthmax_agentsready_delay_msmax_lifetime_msdefault_permissionslanguage

Set one or more editable config fields.

save_presetrequiredrun_idnameoptionaldescription

Capture a live run as a preset.

start_presetrequirednameoptionalrun_nameworking_dir

Start a run from a preset.

delete_presetrequiredname

Delete a preset.

Hosts and skills

attach_hostoptionalkey

Attach the MCP to one host CLI, or all.

detach_hostrequiredkey

Detach the MCP from one host CLI.

install_skillsoptionalstatus

Install the ReevesAgents skill for skill-aware CLIs, or report where it is.

Two resources carry the provider catalog and a usage guide: reevesagents://providers, reevesagents://guide

Configuration

Settings live in ~/.reeves/config.json. Change one with reevesagents config <key> <value> or from the settings screens. reevesagents config language tr switches the interface to one of its ten languages.

Fields

peek_interval_ms5000

Saved polling-interval setting. Current UIs use fixed refresh intervals.

peek_lines10

Saved output-line setting. Current screens and CLI/MCP reads use their own line counts.

max_agents100

Per-run live-agent cap enforced by MCP spawn. Other launch surfaces do not enforce it.

max_depth5

Saved recursion-depth setting. The current run engine does not enforce it.

ready_delay_ms5000

Wait for a fresh CLI to settle before pasting a prompt.

max_lifetime_ms0

Reap agents older than this. 0 turns it off.

default_permissionsask

Default permission mode for new agents: ask or skip.

languageen

UI language, one of the ten locale codes.

State on disk

~/.reeves/
  config.json
  presets/<name>.json
  runs/<run-id>/
    run.json
    agents/<agent-id>.json
    approvals/<approval-id>.json
  history/runs/<run-id>.json

Environment overrides

REEVES_REGISTRY

Override the state root (default ~/.reeves).

REEVES_CONFIG

Override the config file path.

REEVES_NO_TMUX_WRAPPER=1

Do not auto-wrap the TUI into a tmux session.

REEVES_NO_COLOR=1

Disable TUI colors.

ALLOW_DESTRUCTIVE=1

Allow destructive commands without --yes.

REEVES_DOCTOR_SKIP_PROVIDER_COMPAT=1

Skip the provider-compatibility doctor check.

Providers

25 CLIs, each with a short id you use in a spawn spec. Model lists are suggestions, and any model id you pass goes straight to the provider.

  • Claude CodeMCP host

    Anthropic's coding agent for the terminal

    id ccbinary claude

    suggested models fable, opus, sonnet, haiku

  • Codex CLIMCP host

    OpenAI's coding agent CLI

    id codexbinary codex

    suggested models gpt-5.6-sol, gpt-5.5, gpt-5.4-mini

  • OpenCode CLIMCP host

    Open-source coding agent for the terminal

    id opencodebinary opencode

    suggested models opencode/claude-sonnet-4-6, opencode/claude-opus-4-8, opencode/gpt-5.2-codex, anthropic/claude-opus-4-8, anthropic/claude-haiku-4-5, openai/gpt-5.5

  • HermesMCP host

    Nous Research's agent CLI

    id hermesbinary hermes

    suggested models anthropic:claude-opus-4-8, anthropic:claude-fable-5, openai-codex:gpt-5.5, openai-codex:gpt-5.4-mini

  • Kimi CodeMCP host

    Moonshot AI's coding CLI

    id kimibinary kimi

    suggested models kimi-code/k3, kimi-code/kimi-for-coding, kimi-code/kimi-for-coding-highspeed

  • DeepSeek CLI

    DeepSeek's coding CLI

    id deepseekbinary deepseek

    suggested models deepseek-v4-pro, deepseek-v4-flash

  • Pi

    A minimal terminal coding harness

    id pibinary pi

    suggested models sonnet, opus, deepseek/deepseek-chat, openai/gpt-4o

  • Qwen CodeMCP host

    Alibaba's coding agent CLI

    id qwenbinary qwen

    suggested models qwen3-coder-plus, qwen3-coder-flash, qwen3-coder-next

  • Aider

    AI pair programming in the terminal

    id aiderbinary aider

    suggested models sonnet, opus, haiku, deepseek/deepseek-chat

  • Gemini CLI

    Google's open-source terminal agent

    id geminibinary gemini

    suggested models gemini-3.1-pro-preview, gemini-3.5-flash, gemini-3.1-flash-lite

  • Cline CLI

    Autonomous coding agent CLI

    id clinebinary cline

    suggested models anthropic/claude-sonnet-4-6, anthropic/claude-opus-4-8, google/gemini-3.1-pro

  • Crush

    Charm's coding agent for the terminal

    id crushbinary crush

    suggested models chosen inside the CLI

  • Goose

    Open-source general-purpose AI agent

    id goosebinary goose

    suggested models claude-sonnet-4-6, claude-opus-4-8, deepseek-v4-pro

  • Kilo Code

    Open-source coding agent and CLI

    id kilobinary kilo

    suggested models kilo/~anthropic/claude-sonnet-latest, kilo/~anthropic/claude-opus-latest, kilo/~google/gemini-pro-latest, kilo/~openai/gpt-sol-latest

  • Letta Code

    Coding agent with long-term memory

    id lettabinary letta

    suggested models zai-org/GLM-5.3-max, moonshotai/Kimi-K2-Instruct, Qwen/Qwen3-Coder-480B-A35B-Instruct

  • Mistral Vibe

    Mistral's coding agent CLI

    id vibebinary vibe

    suggested models chosen inside the CLI

  • Grok Build

    xAI's coding agent CLI

    id grokbinary grok

    suggested models grok-4.6, grok-4.5, grok-4.7

  • Kiro CLI

    AWS's agentic coding CLI

    id kirobinary kiro-cli

    suggested models claude-sonnet-4.6, claude-opus-4.7, claude-opus-4.6

  • Factory Droid

    Factory's coding agent CLI

    id droidbinary droid

    suggested models chosen inside the CLI

  • Qoder CLI

    Qoder's agentic coding CLI

    id qoderbinary qodercli

    suggested models auto, performance, lite

  • Warp

    Warp's agent in the terminal

    id warpbinary warp

    suggested models chosen inside the CLI

  • Junie CLI

    JetBrains' coding agent CLI

    id juniebinary junie

    suggested models sonnet, opus, gpt-sol

  • GitHub Copilot CLI

    GitHub's coding agent CLI

    id copilotbinary copilot

    suggested models auto, gpt-5.4, claude-opus-4.8

  • Cursor CLI

    Cursor's coding agent CLI

    id cursorbinary cursor-agent

    suggested models gpt-5, sonnet-4-thinking, claude-opus-4-8

  • Rovo Dev CLI

    Atlassian's coding agent CLI

    id rovobinary acli

    suggested models chosen inside the CLI

  • --effort sets reasoning effort for Claude Code and Codex. --auth-mode api-key applies to Claude Code.
  • Pass any other launch flag with --extra-args, for example Claude Code's --remote-control.
  • Keep permission prompts enabled. --skip disables them in the coding tools.

Agent tools

Four helper CLIs an agent can run from its own shell. They cannot take a chat task, so they are not spawnable providers. reevesagents tools lists them with install and sign-in status.

  • Perplexity CLIpplx

    Web search and page snippets from Perplexity, as JSON.

  • Google Colab CLIcolab

    CPU, GPU and TPU runtimes on Google Colab: run scripts and notebooks remotely.

  • Hugging Face CLIhf

    Download and upload models and datasets, run jobs on the Hugging Face Hub.

  • Ollamaollama

    Run local or Ollama cloud models from the terminal.

Troubleshooting

Start with reevesagents doctor. It usually names the problem.

tmux is missing

Install it with brew install tmux or sudo apt install tmux, then run doctor again.

A provider does not show up

The CLI is usually not installed or not signed in. Fix that, then check with reevesagents providers.

The Web UI says packages are missing

The live terminal needs the optional ws and @lydell/node-pty modules. Reinstall globally with npm or pnpm to get them back.

Port 8080 is taken

The Web UI moves to the next free port and prints the URL. Pick one yourself with reevesagents web --port 8090.

Native Windows

Run the full product in WSL. For Agent Control only, reevesagents-win runs natively as an MCP server.

Facts you can check

Downloads cover 2025-04-14 to 2026-10-14. GitHub and npm figures checked on 2026-10-15; downloads count package fetches.

Last updated: