Eggify: the AI workstation for writing code

What it actually is

A workshop for your coding agents

Eggify runs in your browser and on your machine. AI agents write, run and test code in your repos while you watch, steer and approve.

localhost:3333 · Agents

YouThe login test fails since the token refactor. Fix it and make sure the auth suite passes.

AgentLooking at the session code first.

Read internal/auth/session.go

Edit internal/auth/session.go +4 −2

-  row := store.Find(token)
+  row := store.Find(hashToken(token))

Bash go test ./internal/auth/...

ok   internal/auth   0.42s   38 passed

AgentFixed: sessions are looked up by the hashed token again. All 38 auth tests pass.

Real repositoriesAgents work in your own folders and git branches, with Bash, Read, Write and Edit.
You stay in chargeApprove risky tool calls, keep sessions sandboxed, interrupt or switch model at any time.
Nothing hiddenEvery prompt, diff, command, token and cent is on screen and saved.

The film

Sixty seconds, one egg

Eggify in a minute: real screens, a posh announcer and, inevitably, the foot.

The programme

Six sketches, one egg

Everything a developer needs to keep a troupe of AI coding agents productive, accountable and, broadly speaking, under control.

Sketch No. 1

Five identical clerks in bowler hats typing on typewriters

Parallel agent sessions

The Ministry of Many Sessions

Run several agent sessions side by side, each with its own working directory, git branch and permissions, and full tool support: Bash, Read, Write, Edit and the rest.

Sketch No. 2

A steam-powered contraption covered in gauges, printing ticker tape

Live costs & metrics

The Gauges of Inevitable Truth

Live token usage, costs, tool executions, git status and running processes, streamed over WebSockets the moment they happen.

Sketch No. 3

An Edwardian switchboard operator plugging in cables

Any model, any provider

The Switchboard

Claude, OpenAI Codex, DeepSeek, GLM, Kimi or any Anthropic-compatible endpoint. Move a live session to another model without starting over; the conversation comes along.

Sketch No. 4

An egg in a bowler hat sitting inside a gilded birdcage

Sandbox & permissions

The Gilded Cage

Sandboxed sessions with an egress allowlist, tool-by-tool approvals, protected secrets and workspace trust. And YOLO mode, for when you simply want it done.

Sketch No. 5

A gentleman shouting into a gramophone horn, letters flying out

Local voice input

The Shouting Horn

Speak your prompt instead of typing it. Speech-to-text with Parakeet runs locally, no cloud API and no API key.

Sketch No. 6

Session handovers

The Hand-Over

Pass a session's context, history and notes to a fresh agent with a single-use token, and carry on exactly where the last one left off.

“I came here for an argument with my agents, and got a perfectly reasonable audit trail.”

A satisfied operator, probably

Getting started

Hatching instructions

Three steps. No ministry approval required.

The desktop app

Eggify for Mac, server included. Drag it to Applications, open it, and it keeps itself up to date.

Download for macOSVersion 2.3.1 · Apple Silicon · macOS 13.4+

Or build it yourself:

Build the binary
Eggify is a single Go binary with the dashboard built in. Build it from source with just build (needs Go and Node).
Create your admin
Run ./eggify. It opens https://localhost:3333, where you choose a username and password.
Release the agents
Add a provider, pick a project and start a session. The egg in the corner tips his hat when they get to work.

terminal.txt

# fetch the source
$ git clone https://github.com/schlunsen/eggify
$ cd eggify

# build (frontend + Go binary)
$ just build

# and now for something completely different
$ ./eggify

macOS & LinuxSingle binarySQLite, no setupHTTPS out of the boxiOS companion appMIT licensed