Vibeleaderboard
Index / agent
Visit github.com
Category
AI Agents
Rank
No. 1381Tools index

Previous survey · No. 1394 ·

Pricing
Open Source
Type
AGENT
Builder
nexu-io
GitHub
115 stars
Latest release
v0.13.0
Date

About

Run AI coding agents as an autonomous dev team across all your GitHub repos: plan, review, fix, and ship on a loop.

What it does

Looper runs locally as a daemon with a companion command-line interface. Repository issues, assignments, labels, pull requests, and review threads become its workflow state. Separate planner, reviewer, fixer, and worker roles operate in isolated Git worktrees, handing work between stages until a pull request is ready for human review and merge.

Why it's ranked here

The design has unusually clear control boundaries for unattended coding work. Labels gate transitions, every role has an explicit exit condition, and operators can pause, inspect, stop, resume, or take over loops. It also avoids tying the workflow to one coding agent. The cost is substantial local machinery and dependence on disciplined forge state.

What's good

Each concurrent loop gets its own Git worktree, reducing collisions across issues and repositories. GitHub remains the workflow authority instead of a separate task database. Reviewer and fixer roles can alternate until actionable threads are handled. A single daemon watches multiple projects, while manual commands support planning, reviewing, implementation, diagnostics, logs, and targeted pull request takeover.

Tradeoffs

This is not a hosted service. The daemon, forge credentials, Git checkout, and supported agent command must remain available locally. Default operation polls for changes. GitHub has full support, while Forgejo support covers a narrower stated set of planning, working, reviewing, and compatibility flows. Automatic progress also depends on labels, assignments, review state, and agent output meeting expected conventions.

How to use it well

Use it when your team already treats issues, pull requests, labels, and reviews as the operating system for development. Start with one pull request takeover, inspect the review and fix cycle, then register more repositories if the controls fit. Keep human merge review as the final boundary. It does not replace a hosted control plane or a general external task tracker.

Technical notes+

The module in go.mod targets Go 1.22 and depends directly on Cobra, SQLite, TOML, terminal, progress-bar, and YAML packages. internal/cliapp/app.go defines the broad looper command surface. internal/api/server.go hosts an HTTP server, while internal/api/handler.go exposes a versioned API and dashboard-oriented configuration controls. internal/agent/executor.go runs external agent processes, persists execution observations, limits captured output, supports native session resume, and filters vendor-specific arguments. internal/agent/prompt.go requires a structured completion marker. internal/agent/codex_jsonl.go translates Codex JSONL events for session IDs, command status, output, and terminal state. Tests are colocated in internal/api/api_test.go, internal/api/handler_test.go, and internal/agent/prompt_test.go.

Observed

License
MIT
Primary language
Go 1.22
Packaging
Ships separate looper CLI and looperd daemon binaries, with an install script and source installation documentation.
Interfaces
Command-line interface, local HTTP API, and local operator dashboard.
Platform support
Fast-path binaries are documented for macOS arm64 and Linux amd64.
Forge support
GitHub is fully supported; Forgejo supports planner, worker, native review requests and reviews, plus summary-comment compatibility flows.
Agent vendors
Documented adapters include OpenCode, Claude Code, Codex, Cursor CLI, and Grok Build.
Storage
The Go module directly depends on SQLite through go-sqlite3.

Read from README.md, go.mod, internal/api/doc.go, internal/agent/doc.go, internal/api/server.go, internal/cliapp/app.go, internal/api/handler.go, internal/agent/prompt.go, internal/api/api_test.go, internal/agent/executor.go, internal/api/handler_test.go, internal/agent/codex_jsonl.go, internal/agent/prompt_test.go, internal/agent/spawn_owner.go, internal/api/browser_guard.go.

What it can do

  • Plan software development tasks

    GitHub repository and project requirementsDevelopment plan with tasks and priorities

  • Review code changes automatically

    Code commits and pull requestsCode review feedback and suggestions

  • Fix bugs and issues in code

    Bug reports and code with issuesFixed code with resolved issues

  • Deploy code changes to production

    Reviewed and approved code changesDeployed application updates

  • Monitor multiple GitHub repositories

    GitHub repository URLs and access permissionsContinuous monitoring and management across repos

  • Execute development workflows autonomously

    Development workflow configurationsCompleted development cycles without manual intervention

Tags

ai-codingautonomousgithubagents

Tech Stack

Go

Comments (0)

No comments yet

Editorially curated, with community endorsements as a secondary signal. Corrections welcome.