AGENTIC CODING LOOPS
Orchestrate a whole team of AI coding agents from one place, with you in the loop. Now with five ways to point AI at your code, a private brain for every project, a voice, and permission approvals from your phone.
60-day free trial, no credit card. $10/month after. See what's new
● Release · v3.2.1 · out now
Graph Mode, a fifth loop mode that runs a job as a dependency graph of worker nodes, and a phone companion app that pairs with the desktop over your local network so you can approve permission prompts from your phone. Plus copyable code blocks in chat, bounded logging, and steadier instance startup. Fully backwards compatible.
Graph Mode is an early preview and ships switched off: turn it on per user under Settings, Agents, Loop Modes. Pair a phone under Settings, Mobile. Built-in providers remain Claude Code and Codex; OpenCode plugs in as a custom provider. The knowledge graph and voice remain optional, off by default, and stay on your machine. The AI coding agents connect to your chosen model provider to run. See Terms for details.
Alongside the original RALPH loop, four autonomous modes, each a different way to go from intent to verified work. Five modes, equal footing.
The autonomous loop that drives a spec to done with fresh-context workers, start, pause, resume, stop.
Turn a one-line idea into a reviewed, approved plan, requirements, design and tasks, before a single line of code.
Chase a measurable target: try an idea, measure, keep what improves the number, revert what doesn't.
One agent builds; an independent reviewer grades each step against your criteria and sends it back until it passes.
Describe the work and the repository to the Graph Coordinator and it proposes a graph of parallel worker nodes with edges between them, written to a config the app picks up and put on the board for you to approve. Then watch it run.

Settings, Agents, Loop Modes: enable Graph, set default concurrency, auto-approve runs
A private, per-project knowledge graph built from your codebase. Local and private, off by default.

A cross-linked wiki of your codebase
Build a wiki. Echorb maps your project into linked pages, entities, concepts and decisions, you can browse and cross-reference.

Ask in plain English, answers cited to real files
Ask it anything. Pose a question in plain language and get an answer grounded in your actual files, every claim cited.
An Android companion app pairs with the desktop over your own network. See every instance and its state, and answer permission prompts from the couch, the kitchen, or the other office.
A few more views from inside Echorb, plus two-way file sharing in chat and a broad production-readiness pass.
Three steps from install to launching your AI team.
Install a supported coding agent (Claude Code or Codex), then point Echorb at it on first run. Other CLIs, such as OpenCode, plug in as custom providers. Add Node.js for MCP server support.
Download for Windows, macOS or Linux. Your existing projects and work carry straight over.
Spin up agents, pick a loop mode, and coordinate everything from one dashboard, or approve from your phone.
Three platforms, one release. 60-day free trial, no credit card. $10/month after.
Version 3.2.1 · Windows 10/11 (x64), macOS 14+, Linux with GTK 3 (x86_64). Fully backwards compatible, existing work keeps running.
Important: Git must be installed and functional in your terminal (PowerShell for Windows, Terminal for macOS / Linux). Download Git: Windows | macOS | Linux
Note: these are newly published apps, so your system may warn you on first launch. On Windows, choose "More info" then "Run anyway" at the SmartScreen prompt. The macOS build is signed and notarized, so it opens normally. On Linux, make the AppImage executable (chmod +x) before running.
Cross-platform desktop. Bring your own coding agent.
System Requirements:
System Requirements:
System Requirements:
parecord, pactl) for voiceffmpeg on PATH for capture