Pi Boomerang is an extension for the Pi coding agent that reduces how much task history carries into later prompts. It's for developers whose bug fixes and refactoring jobs fill the conversation with file reads, edits and test output. After a task finishes, it replaces that history in future model context with a handoff summary. File changes stay on disk, and the full history remains accessible in Pi's session tree.
The summary includes the outcome, changed files, relevant reads, validation commands and detected failures. It appears in chat and goes directly into the coordinating agent's follow-up turn, so the agent can continue with a record of the work rather than rereading the whole task.
Pi Boomerang can run prompt templates in sequence, with separate model, skill and reasoning settings for each step. A chain returns one summary at the end. Repeated passes carry forward earlier summaries and retain file edits, which suits tasks that need several rounds of review or refinement. Shared summary points let separate tasks accumulate context together.
The extension handles summarization outside the worker's task, rather than asking the worker to monitor token limits and decide when to shorten its own context. An optional agent-callable tool lets Pi initiate tasks or manage summary points itself; that capability is disabled by default.
Summaries have limits: the extension extracts details from recorded tool calls and results, so it can miss the meaning behind a change.
Claim this page and we'll verify you by hand. Pi Boomerang gets the verified badge, and you can upgrade the listing to be featured on localhosted. Proud to be listed? Put our badge on your site.
Want more people to find Pi Boomerang?Promote it
Something wrong or outdated on this page?
162Updated 2 months ago
#Code execution#MCP#Multi-agent workflows
Forge Orchestrator coordinates Claude Code, Codex CLI, and Gemini CLI when they work on the same repository. It's for developers who use multiple AI coding assistants and need them to share decisions, respect file ownership, and follow the same specification.
1.2KUpdated 4 weeks agoMIT
#Agent Skills#Code execution#Git integration
4.4KUpdated 1 day agoMIT
macOS · Windows · Linux · Web · JetBrains#Agent Client Protocol#Code execution#Git integration
15Updated 1 week agoMIT
macOS · Linux · Web#Agent Skills#Code execution#Git integration
126Updated 1 month agoApache-2.0
#Agent Skills#Code execution#MCP
106Updated 2 weeks ago
macOS · Windows · Linux · Web#Agent Skills#Code execution#Git integration
Albert coordinates Claude Code agents for developers who want to hand off a coding goal across several tasks. Its harness and dashboard run locally on Windows, macOS or Linux, and it requires an installed, signed-in Claude Code account. It assigns work to Haiku, Sonnet or Opus according to the task and continues until the goal is complete or it reaches a requested review checkpoint.
Pi Subagents adds specialized AI agents to pi for developers who want to split coding work across separate sessions. Each agent can have its own model, prompt and tool permissions. The extension runs within pi and is open source under the MIT license.
gptme is a self-hosted AI agent that works directly in your terminal, with access to your files and installed tools. It's for developers who want a coding assistant in their own environment, and people who want an agent for data analysis or other knowledge work. The software is free under the MIT license and doesn't require a gptme account.
Agents Studio is a self-hosted app for developers who want Claude Code to work on their repositories while they're away. It runs on your machine, with browser and terminal interfaces, and supports macOS and Linux. The software is open source under the MIT license. Claude Code must be installed and signed in; sessions can also use Cursor's signed-in agent.
aionrs is a Rust command-line AI agent for developers who want an assistant to work directly with local files and shell commands while choosing their own model provider. It can inspect a project, edit code, search files and view images, then carry out tasks across multiple tool calls.