Skip to content
OpenWaggle
OpenWaggle
Open Source Desktop Coding Agent ALPHA

The coding harness that

lets your agents waggle.

In nature, bees don't solve problems alone — they perform a waggle dance to share what they've found. OpenWaggle works the same way: two AI agents collaborate in structured turns until they converge on solutions no single model would reach.

Product Screenshot App UI with Waggle Mode active
Works with 6 providers
AnthropicOpenAI Gemini Coming soonGrok Coming soonOpenRouter Coming soonOllama Coming soon
Powered by TanStack AI SDK ↗
HOW WAGGLE MODE WORKS

Two models collaborate like bees in a hive

Just as forager bees perform a waggle dance to share what they've found, OpenWaggle orchestrates two AI agents in structured turns until they converge.

1

Configure your team

Pick two models from different providers. Assign roles, strengths, and collaboration rules. Save as a reusable preset.

2

Agents collaborate

They take structured turns — reading files, running commands, challenging each other's assumptions. You watch in real time.

3

Consensus reached

When agents converge on a solution, a synthesis step combines their findings into a clear recommendation with agreed points and open questions.

CORE FEATURES

Everything a coding agent needs

Full coding agent

Read, write, and edit files. Run shell commands. Browse the web. Ask clarifying questions. All with approval-based execution — you stay in control of every destructive operation.

Git-native workflow

Live diff stats in the header. Branch management from the picker. Stage, commit, and review changes in a side-by-side diff panel — all without leaving the conversation.

Infinitely extensible

Custom skills add specialized knowledge to the agent. MCP servers connect external tools — from Playwright browser automation to Chrome DevTools. Per-project AGENTS.md files scope instructions to specific paths.

BUILT FOR DEVELOPERS

Every detail considered

Local-first privacy

Conversations, API keys, and voice data stay on your machine. No telemetry. No cloud sync.

Voice input

Local Whisper transcription. Speak your instructions — no audio leaves your machine.

Plan mode

The agent outlines its approach before executing. Review, revise, and approve the plan interactively.

Rich attachments

Drop text files, PDFs, and images into the conversation. Content is extracted automatically with OCR for images.

Built-in terminal

Full PTY terminal emulation with xterm.js. Toggle with a shortcut — run commands without leaving the app.

Execution modes

Default mode requires approval for writes and shell commands. Full access mode executes immediately. You choose the trust level.

Ready to waggle?

Download OpenWaggle and let two AI models solve your next problem together.

Free and open source · macOS, Windows, Linux