uta: A CLI Agent Orchestrator for Parallel AI Task Management

Introduction

uta, short for 'unleash the agents', is a powerful command-line interface (CLI) agent orchestrator. It specializes in breaking down complex goals into independent, parallel subtasks and distributing them to any installed coding-agent CLI, such as Claude Code or Gemini CLI. uta then synthesizes the results and meticulously records every step as a queryable trajectory, providing full transparency and auditability.

Why use and added value

While individual agent CLIs excel at single back-and-forth conversations, uta operates as a crucial layer above, offering significant advantages:

• Parallelism: A built-in planner decomposes goals into independent subtasks, allowing them to run concurrently across multiple providers, significantly speeding up execution.
• Cross-provider Compatibility: Each subtask can target a different CLI, leveraging the strengths of various models, for example, Gemini for large context or Claude for delegation-heavy edits. Custom YAML descriptors allow integration with virtually any other agent.
• Replayable Trajectories: Every prompt, tool call, token count, and final answer is recorded in SQLite. Commands like uta sessions and uta trajectory <id> enable inspection, auditing, or resuming any past run.
• Modes & Profiles: Mission Profiles allow setting policies per task type (e.g., dev, ops, audit, research), defining allowed tools, human-in-the-loop triggers, budget caps, and retrospective cadences.
• Bidirectional MCP: uta can call Model Context Protocol (MCP) servers as tools and can also expose itself as an MCP server, allowing other MCP-aware clients like Claude Code or Cursor to integrate with uta's planner.

Installation

Installing uta is straightforward, with multiple options available:

• Homebrew (macOS / Linux):

  brew install unleashtheagents/tap/uta

• curl | sh:

  curl -fsSL https://unleashtheagents.ai/install.sh | sh

• Go toolchain:

  go install github.com/unleashtheagents/uta/cmd/uta@latest

Examples

Get started with uta through a quick 60-second tour of its core functionalities:

uta hello                                             # what is this, what's installed, what to try next
uta doctor                                            # verify environment
uta providers                                         # list detected agent CLIs
uta run -g "summarize this repo in 5 bullets" -y      # ad-hoc orchestration
uta sessions                                          # list past runs
uta trajectory <id>                                   # the full event timeline
uta resume <id> -g "now turn each bullet into a tweet"
uta discuss -t "monolith or microservices for this repo?"   # two LLMs debate, moderator synthesizes
uta shell                                             # interactive shell: converse with + control agents
uta mission run examples/hello.steer                  # run a steer program (the agent language, v0)

The uta discuss command enables two agents, ideally different LLMs, to debate a topic over several rounds, with a moderator synthesizing agreements, disagreements, and a final verdict.

uta shell provides a persistent, interactive terminal prompt where you can converse with and control agents. It supports slash commands for model control, tool management, and routing turns to specific agents.

uta mission is the v0 interpreter for steer, a programming language specifically designed for agents. It manages agent calls as effects, enforces linear and mandatory budgets, and journals every run to the session store.

Here's a sample output of a fan-out run:

$ uta run -g "summarize this repo in 5 bullets" -y
[uta] planner=claude/gemini worker=claude max-parallel=3
[uta] plan: 3 subtasks
  s1  what does this codebase do?
  s2  how is it structured?
  s3  what's the install path?
[uta] s1 ? running (claude)         ? 1.2s
[uta] s2 ? running (claude)         ? 2.1s   parallel with s1
[uta] s3 ? running (claude)         ? 0.9s   parallel with s1/s2
[uta] synthesize ?? claude          ? 0.7s
- A CLI for orchestrating multiple agent CLIs in parallel.
- ...

[uta] session 4f6c... status=completed  tokens=12,430  approx $0.04

Links

For more detailed information and to contribute, visit the official GitHub repository: uta GitHub Repository