juggler  by juggler-ai

Visual workbench for hands-on AI coding agent control

Created 1 month ago
516 stars

Top 60.0% on SourcePulse

GitHubView on GitHub
Project Summary

Summary

Juggler provides a visual, hands-on workbench for AI coding agents, targeting users who need granular control and visibility over LLM interactions. It replaces linear chat transcripts with an inspectable, editable tree structure, offering a GUI for deep navigation and modification of AI sessions, thereby enhancing developer productivity and understanding.

How It Works

Juggler employs a native desktop application and browser clients that connect to a local webserver, treating AI sessions as synchronised Yjs documents. Its core innovation lies in representing conversations as editable trees rather than linear logs. Users navigate and inspect tool calls, context, and sub-threads via a Miller column interface. Functionality is highly extensible through JavaScript plugins for context items, LLM strategies, and slash commands, allowing users to inspect, fork, or replace core components.

Quick Start & Requirements

Pre-built binaries are available via juggler.studio or the Releases page. Installation involves downloading macOS (.dmg), Windows (.exe), or Linux server binaries. The desktop app and server are bundled. Official binaries support WAN access; source builds are local/LAN only. Refer to docs/headless-linux.md for headless setups.

Highlighted Details

  • Editable Tree Session: Replaces linear chat history with a navigable, branchable, and editable tree structure.
  • Miller Column UI: Provides visual navigation and inspection of session items, properties, and nested sub-threads.
  • Extensible Plugin System: Core components (context items, commands, strategies) are JavaScript plugins, allowing deep customization.
  • Multi-Client Architecture: Enables synchronized access to a single session from the desktop app, browsers, or different machines.
  • Broad Model Support: Integrates with Claude Code, OpenAI, Gemini, Ollama, OpenRouter, Z.AI, Deepseek, and others.

Maintenance & Community

The README does not detail specific community channels (e.g., Discord, Slack) or maintainer information. Contribution guidelines are available in CONTRIBUTING.md, and security issues should be reported via SECURITY.md.

Licensing & Compatibility

The core Juggler application code is licensed under AGPLv3, imposing copyleft obligations on distributed or hosted modified versions. The extension SDK and bundled extensions use the permissive Apache-2.0 license, allowing closed-source extensions without copyleft requirements. Commercial use or closed-source integration requires contacting the author for licensing.

Limitations & Caveats

The AGPLv3 license necessitates careful consideration for commercial or closed-source integration due to its strong copyleft provisions. Builds from source are restricted to local and LAN access, with WAN capabilities reserved for official binaries.

Health Check
Last Commit

1 day ago

Responsiveness

Inactive

Pull Requests (30d)
19
Issues (30d)
38
Star History
519 stars in the last 30 days

Explore Similar Projects

Feedback? Help us improve.