Coding agents crossed from autocomplete to running whole tasks end to end. They run long, they run in parallel, and increasingly they run unattended. binigma is the layer that lets one person run a lot of them — on hardware they own.
binigma is a self-hosted, installable app that points at a small server running on a machine you own. That server spawns Claude Code, Codex and raw terminal sessions in any project directory, and lets you drive — and orchestrate — all of them from a fast, native-feeling interface on any device.
It is a developer tool with the polish of a consumer app. Your source never leaves your box, and there is no LLM key to store, because it runs on the CLI subscription you are already logged into.
Each agent is an integration, not a competitor. Claude Code and Codex today, more as they ship. One human, one console, a fleet of autonomous engineers.
| Runs on | Your Apple Silicon Mac or a Raspberry Pi 5 |
| Agents | Claude Code · Codex · plain terminal |
| Access | Installable PWA on your phone, plus desktop web |
| Orchestration | One manager, many subagents, each in its own worktree |
| Your data | Stays on your own box |
| Always on | Sessions survive a closed lid |
Yesterday you left your laptop open overnight so Claude Code could finish a task while you slept.
Tomorrow you leave the office and your engineering team of agents keeps working — and you review, steer and merge from your pocket.
That second sentence is only true if the machine running the agents never goes to sleep, and if the thing you hold in your hand is good enough to actually review work on. Those are the two problems binigma solves.
Promote any session to a manager and it stops writing code — it starts running the project. It scopes the work into issues, spawns a dedicated subagent per task, and drives each one through your real engineering pipeline, in parallel.
Turns a rough goal into well-formed, verified GitHub issues, one crisp task at a time.
Every issue gets its own git worktree off the latest master, so parallel agents never collide on the same files.
A dedicated subagent investigates, forms a plan, and implements the change end to end.
Runs the build, the typecheck and the test suite, and iterates until it is green.
Independent reviewer agents adversarially check the diff before anything is trusted. Findings converge over rounds rather than in one pass.
Rebases each branch onto the freshest master and resolves the drift as sibling work lands.
Orders dependent work and gates each merge. You approve the boundaries that matter.
Master is never written autonomously. You steer from the top and it does the rest, but every subagent works in its own isolated worktree and the merge into your main branch is a boundary you approve.
You manage the manager; it manages the fleet. A single instruction from your phone — "build OAuth login" — becomes a manager agent that scopes, dispatches and reviews: a frontend subagent, a backend subagent, an API subagent, QA, a security review. Five pull requests come back, one per task, and you gate the merges.
binigma is not a browser tab you leave open — it is a service on your always-on machine. Kick off work at your desk, shut the lid, and every session keeps building.
Open your phone on the train and pick up exactly where you left off, with full history intact. Sessions reconnect automatically: no cold starts, no lost context. Web Push, Telegram and voice ping you the moment an agent needs you.
Start Claude Code, Codex or terminal sessions; read their reasoning, reply, approve plans and answer the questions they raise.
Worktrees, branches, pull requests and issues, inline. Review diffs with +/− counts, open a full xterm, rebase onto master.
Promote a session to a manager that dispatches, coordinates and reviews other agents in a shared forum.
Claude Code and Codex today, more on the way — plus custom skills and prompts that accrue to you rather than to a vendor.
Speak your intent and hear replies read back — a genuinely hands-free flow when you are away from a keyboard.
Browse the working tree, edit files in a real editor with go-to-definition, gate CI, approve and squash-merge pull requests — all without a laptop.
Install it to your home screen and you get a mobile-first shell with a floating glass tab bar: Home, Sessions, Repos, Files and Skills, all one thumb-reach away. It is a real app, not a shrunken website.
The same console swaps to a desktop layout: a resizable sessions rail, a multi-tab workspace, and an expanding side panel for side-by-side diff review. A full xterm.js terminal, keyboard-first, and everything the PWA does.
The reason binigma is self-hosted is not ideology — it is that agents need real access to your source, your git remotes and your shell. The safest place for that is a machine you already trust.
Voice is off by default and uses your own transcription key. When you turn it on you can dictate your intent straight into the composer, hear replies read back while you walk or drive, and have a whole issue or pull request read aloud on the go.
Running binigma on your own machine is free — single-user or multi-user, no account, no seat count, and not a trial. The software is the distribution; charging for it would only slow that down.
What will be paid, later, is the part that costs real money to run: a hosted box we operate for you if you would rather not run hardware, and usage credits for the work that runs through it. Self-hosting stays free regardless.
You bring your own Claude or Codex subscription either way — binigma never resells model access.