Agent Orchestrator

Executive Summary

Agent Orchestrator manages fleets of AI coding agents working in parallel on your codebase. Each agent gets its own git worktree, its own branch, and its own PR. When CI fails, the agent fixes it. When reviewers leave comments, the agent addresses them. You only get pulled in when human judgment is needed.

Born inside Composio in February 2026, the repository was transferred from ComposioHQ to creator Prateek Karnal's AgentWrapper GitHub account by June 2026 as Composio refocused on MCP infrastructure. Development has not slowed: 7,503 stars, 1,020 forks, nightly releases through June 8, 2026, and a last push on June 9, 2026.

---

Product Overview

Agent Orchestrator solves the coordination problem: running one AI agent in a terminal is easy — running 30 across different issues, branches, and PRs is hard.

Without orchestration, you manually create branches, start agents, check if they're stuck, read CI failures, forward review comments, track which PRs are ready to merge, and clean up when done.

With Agent Orchestrator: ao spawn and walk away. The system handles isolation, feedback routing, and status tracking. You review PRs and make decisions — the rest is automated.

Key Capabilities

Workflow

ao spawn my-project 123

1. Workspace creates isolated git worktree with feature branch
2. Runtime starts tmux session (or Docker container)
3. Agent launches Claude Code (or Codex/Aider) with issue context
4. Agent works autonomously — reads code, writes tests, creates PR
5. Reactions auto-handle CI failures and review comments
6. Notifier pings you only when judgment is needed

---

Technical Architecture

Plugin Architecture

Seven slots, every abstraction swappable:

Reaction System

reactions:
  ci-failed:
    auto: true
    action: send-to-agent
    retries: 2
  changes-requested:
    auto: true
    action: send-to-agent
    escalateAfter: 30m
  approved-and-green:
    auto: false  # flip to true for auto-merge
    action: notify

Key Technical Details

---

Strengths

• Parallel isolation — Each agent gets its own git worktree, branch, and PR; no merge conflicts during development
• Autonomous CI handling — Failed builds automatically routed to agent with logs; configurable retries
• Review automation — Reviewer comments forwarded to agent; escalation after timeout
• Agent-agnostic — Claude Code, Codex, Cursor, Aider, OpenCode, KimiCode supported; plugin architecture for others
• Real community traction — 467 stars at launch grew to 7,503 by June 2026, with 1,020 forks and nightly releases
• Proven pedigree — Built at Composio (18K-star platform); the system famously built itself, with agents merging 80+ of their own PRs in the first 8 days

---

Cautions

• Lost corporate sponsor — Composio transferred the repo to the creator's personal AgentWrapper account by June 2026 as it refocused on MCP infrastructure; the project no longer has a company behind it
• Bus factor — Primary maintainer is one person (Prateek Karnal); 945 open issues as of June 2026 suggest triage is straining
• No stable 1.0 — Still on v0.9.x nightlies as of June 2026
• tmux/Docker required — Default runtime needs tmux; Docker option available but adds complexity
• GitHub-first — Linear support via plugin, but GitHub is the primary integration
• No enterprise features — No SSO, audit logs, or air-gapped deployment yet
• Idle gaps — One hands-on reviewer observed agents sitting idle 90+ seconds waiting on tool approval in multi-agent runs
• CLI-focused — Dashboard exists but CLI is primary interface

---

What Developers Say

Independent community discussion is thinner than the star count suggests — the Show HN launch drew only 3 points and one comment, and no substantial Reddit threads surfaced in research. The most useful signals:

"Of every tool I tested, AO is the one that pushes hardest past session management into autonomous PR handling... The lifecycle manager caught three CI failures, pulled the logs, fed them to the agent, and retried twice before escalating one to me."

— Augment Code's hands-on comparison of nine open-source orchestrators (May 2026)

The same review flagged friction: "Agents go idle more than I expected between steps. On a four-agent run, two sessions sat idle for 90+ seconds waiting on tool approval."

From the creator, on why orchestration is harder than session management:

"I run 16-30 Claude Code sessions in parallel using an orchestrator I built... Session management is half the problem. The bigger problem is managing your own context around everything that's going on."

— Prateek Karnal (prateekk77) on Hacker News, March 2026

---

Pricing & Licensing

Licensing: MIT — use commercially, modify freely.

Hidden costs: You pay for underlying agent subscriptions (Claude Max, ChatGPT Pro, etc.).

---

Competitive Positioning

vs Other Orchestrators

When to Choose Agent Orchestrator

• Choose Agent Orchestrator when: You want parallel agents with automated CI/review handling and plugin flexibility
• Choose Gastown when: You need maximum parallelism (20-30 agents) and sophisticated role-based orchestration
• Choose Metaswarm when: You want cross-model adversarial review and enforced quality gates
• Choose Ralph when: You prefer radical simplicity (bash loop until done)

---

Ideal Customer Profile

Best fit:

• Teams running multiple parallel coding tasks
• Projects with CI/CD pipelines that can benefit from auto-fix
• Developers wanting Slack/webhook notifications on agent status
• Users who want to extend with custom plugins

Poor fit:

• Solo developers with single-agent needs
• Enterprise requiring SSO/audit logs
• Teams without GitHub as primary SCM
• Those who prefer GUI over CLI

---

Viability Assessment

The June 2026 picture is a trade-off: the project lost Composio's institutional backing (Composio has refocused on MCP integration infrastructure), but gained 16x the stars and remains under active, near-daily development by its creator.

---

Bottom Line

Agent Orchestrator is the most practical parallel agent orchestrator — focused on the coordination problem without overcomplicating it. The auto CI fix and review handling are genuinely useful; the plugin architecture allows customization without forking. Its 16x star growth since launch (467 → 7,503) validates the approach, but the quiet handoff from ComposioHQ to the creator's personal account changes the risk profile.

Recommended for: Teams running parallel coding agents who want automated CI/review handling and a dashboard to monitor everything — and who are comfortable depending on a creator-led OSS project.

Not recommended for: Enterprise requiring compliance features or vendor accountability, or solo developers who don't need orchestration.

Outlook: Strong position in the orchestrator category on features and adoption; the open question is sustainability. With Composio refocused on MCP infrastructure and TrustClaw, Agent Orchestrator now lives or dies on Prateek Karnal's continued attention — mitigated by MIT licensing, 1,020 forks, and a genuinely active release cadence as of June 2026.

---

Research by Ry Walker Research • methodology