@AISuperDomain: Discovered an ultra-cool open-source project: Farcaster! This is a native, keyboard-first AI coding agent (Coding Agent…
Summary
Farcaster is an open-source, native AI coding agent workspace that integrates multiple AI agents, terminals, and Neovim into a unified UI for developers.
View Cached Full Text
Cached at: 09/13/26, 11:14 PM
🔥 Discovered an ultra-cool open-source project: Farcaster!
This is a native, keyboard-first AI coding agent (Coding Agents) workspace, designed specifically for geeks and hardcore developers!
It’s not meant to replace your existing tools, but rather seamlessly integrates multiple AI agents, terminals, and Neovim into a unified UI.
✨ Core Highlights:
🚀 Seamless Multimodal: Unified management of multiple Agent Harnesses like Claude Code, Cursor, Codex, Pi, and more. ⌨️ Deep Neovim Integration: Each session has its own dedicated Agent | Editor | Terminal, supporting direct code review, selection, and context sending right in Neovim. 🛠️ MCP Toolchain: Built-in workflows (Workgraph), multi-agent collaboration, and code review tools. 🦀 Ultimate Performance: 98.8% written in Rust for a native desktop-grade app experience.
If you’re a heavy Neovim user who relies on multiple AI coding assistants in your daily work, this project is absolutely worth checking out!
https://github.com/behzade/farcaster…
#Rust #Neovim #AICoding #DeveloperTools #OpenSource #GitHub
behzade/farcaster
Source: https://github.com/behzade/farcaster
Farcaster
Farcaster is a native, keyboard-first workspace for running and coordinating coding agents across harnesses, with an embedded terminal and Neovim.
Why
The agent space is changing fast, but switching harnesses shouldn’t mean relearning the interface or leaving your editor behind. I wanted to use different agents and subscriptions with the same controls, and keep Neovim at the center of my workflow.
Is this another agent harness?
No. Farcaster brings your existing harnesses and subscriptions into one UI. It builds around them rather than competing with them.
What does this app give you?
- A native, keyboard-first desktop app.
- An agent, editor, and terminal for each session.
- Integrated Neovim for reviewing code, editing, and sending context to agents.
- Shared tools for coordinating workers and tracking tasks across harnesses.
Neovim
The app embeds a full terminal emulator using libghostty. Each chat session has its own Neovim and terminal tab. I find this a pretty nice way to work with AI as I can think of each session as a separate row of agent | editor | terminal.
The following integrations with Neovim are built into the app:
- Clicking on any changed file opens it in Neovim at the changed line.
- You can open the transcript itself in a temp Markdown file in Neovim to copy/select things.
- You can send messages with context from Neovim (normal mode -> line context, select -> selected context) to the current session, another existing session or a new session to fire off a new task.
- Agents can present their work to be reviewed in Neovim with filenames/lines and context using a built-in tool. (
submit_review)
Tools
Farcaster comes with an optional MCP server (can be turned off in settings) that provides the following capabilities:
-
Workgraph: A set of tools (
workgraph_*) for creating tasks and dependencies that persist beyond any one agent session. -
Worker Tools: Tools for creating and communicating with agent sessions across harnesses, providers, and models.
-
Review: A tool (
submit_review) that lets agents present files, line ranges, and notes for you to navigate and review using Neovim’s quickfix list.
Current status
I have been using Farcaster daily for a while. It’s still early, and bugs/missing features are expected. The current list of supported harnesses is as follows:
- Codex (through codex-cli app server)
- Pi (through Pi RPC)
- OpenCode (through the
opencode2server executable) - Cursor (through Cursor CLI ACP adapter)
- Claude Code (through
claude -p) - Antigravity (through its ACP server)
See the harness feature table for capabilities and usage notes.
In theory, any harness supporting ACP could be integrated through the app’s ACP adapter, but most harnesses support a wider range of features than ACP.
Getting Started
Download the latest build for your platform from the releases page. Builds for macOS ARM64 and x86_64 Linux are currently available.
The macOS app uses ad hoc signing; it is not notarized. If macOS blocks the app because it is not notarized and you trust the download, follow Apple’s instructions to allow it through System Settings → Privacy & Security → Open Anyway.
You’ll also need the agent harness you want to use installed and signed in. To use Neovim in the embedded terminal, it must be available in your environment.
Development
The repository includes a Nix development shell. To build and run from a local checkout:
nix develop
cargo run --locked --bin farcaster
Why build around harnesses?
After a few attempts (tried to fork and extend codex-cli first, then tried building on Pi) I realized that the harness space is very fast moving, with a lot of work and innovation happening in the space. So it would be better to position the app in a way that benefits from said improvements instead of competing with them.
Documentation
Thanks
Thanks to Zed for GPUI, GPUI Component for the UI components, and T3 Code for the UI inspiration.
Similar Articles
Superagent
Superagent is an open-source tool that provides a real computer interface for AI coding agents, featuring a browser and iOS Simulator to streamline developer workflows.
@defileo: If this isn't f**k*ng insane, nothing is. Multica just open sourced a way to run coding agents like real teammates. And…
Multica is an open-source workspace that lets teams assign issues to AI coding agents like Claude Code, Codex, or Cursor, treating them as teammates who comment, report progress, and hand back PRs. It supports 20 agent CLIs and can be self-hosted.
@daniel_mac8: This is one of the coolest open-source AI agent projects I've seen in a while: 'Understand Anything' It's a plugin for …
Understand Anything is an open-source AI agent plugin for Claude Code, Codex, and OpenCode that analyzes codebases and turns them into interactive knowledge bases, enabling users to query and explore their code rather than just viewing structure.
@ms_aifrontiers: Fara1.5 is here! The tech report just landed on arXiv. New SOTA for computer use agents of its size, and it competes wi…
Fara1.5 is a family of native computer use agents trained using the FaraGen1.5 scalable data pipeline. The models achieve new state-of-the-art results on browser-use benchmarks, competing with much larger frontier models.
@AISuperDomain: Obsidian users rejoice: Now AI Agents can truly 'learn' how to operate your knowledge base. kepano has open-sourced obsidian-skills, specifically providing Obsidian support for Agents like Claude Code, Codex, OpenCode, etc. Ob…
kepano has open-sourced obsidian-skills, providing AI Agent skill support for Obsidian knowledge bases, supporting multiple coding agents like Claude Code, Codex, etc., with GitHub reaching over 45K+ Stars.