Rust-native AI delivery studio

Amon Hen

Coordinate Codex, Claude, Gemini, and Linear from one native terminal. Pick the planner, appoint the lead, hand work across providers, fan out same-provider sub-agents, and keep the loop moving until review or green CI.

Codex logo Codex Claude logo Claude Gemini logo Gemini Linear GitHub CI
terminal deliver the Linear project to review or passing CI
cargo install amon-hen
amon-hen --studio --members codex,claude,gemini
Amon Hen Studio live run
Codex logo
Codex model gpt-5.2, effort high, MCP profile repo
planner
Claude logo
Claude sonnet, max effort, allowed tools policy
lead
Gemini logo
Gemini gemini-pro, high effort, settings profile
executor
iterations 2
sub-agents 6
tokens 55k
gate CI
plan execute verify deliver
Rust runtime Cargo workspace, native binary, scriptable stdout, observable stderr.
Provider-native Keep each CLI's auth, MCP, Skills, tools, models, and permissions intact.
Interactive Studio Selectable panes for roles, auth, files, tools, Linear, telemetry, and help.
Delivery loop Poll Linear projects or epics until human review or GitHub CI success.
Studio

A terminal interface that can actually steer the run.

Studio is the control surface for live work. Change roles, tag files, choose auth flows, inspect provider capabilities, watch tool usage, and keep Linear delivery visible while the agents run.

Settings enter Agents 3 Auth tabs Linear on Files @ Tools logs Capabilities MCP Help ?
Settings Lead, planner, handoff, iterations, model and effort per provider.
Capabilities Inherit local provider config or override MCP, Skills, tools, and agents.
Telemetry Token usage, prompt commands, shell tools, sub-agent fanout, and retries.
Linear Project and epic queues, issue workspaces, media attachments, comments, gates.
Providers

Keep the native provider surface. Override only what the run needs.

Amon Hen does not flatten Codex, Claude, and Gemini into a toy abstraction. It lets each CLI keep its auth, tools, Skills, MCP servers, model controls, effort controls, and permissions while Studio exposes the switchboard.

Codex logo Codex planner

model gpt-5.2, effort high, MCP profile repo

Claude logo Claude lead

sonnet, max effort, allowed tools policy

Gemini logo Gemini executor

gemini-pro, high effort, settings profile

--codex-config
--codex-mcp-profile
--claude-mcp-config
--claude-allowed-tools
--claude-disallowed-tools
--gemini-settings
--gemini-tools-profile
Linear delivery

Point it at work and make the loop accountable.

Target projects or epics, isolate each issue workspace, run the planner/lead/executor phases, retry with reconciliation, attach artifacts, and stop at human review or passing GitHub checks.

1
Select

Target projects, epics, teams, states, assignees, labels, or explicit issue IDs.

2
Plan

Use a dedicated planner and lead to break each issue into provider-specific phases.

3
Build

Create isolated workspaces, run tools, fan out sub-agents, test, and reconcile.

4
Deliver

Comment progress, attach media, open or update GitHub work, and wait for review or CI.

terminal autonomous delivery
amon-hen \
  --deliver-linear \
  --linear-project ENG \
  --linear-until-complete \
  --linear-completion-gate review-or-ci \
  --members codex,claude,gemini \
  --planner codex \
  --lead claude \
  --team-work 2
Run it

Install the Rust binary and use your existing provider auth.

Amon Hen shells out to the local provider CLIs you already use. Keep them authenticated, then run Studio or script the exact provider team you want.

terminal native Rust
cargo install amon-hen
amon-hen --auth-status --capabilities-status
amon-hen --studio --members codex,claude,gemini
amon-hen --members codex,claude,gemini --team-work 2 "Inspect this repo"
Links

Everything useful, one click away.