Many agents, one cockpit
Connect Codex and Claude Code today, Gemini soon. Pick which one drives, switch mid-task, and carry on where you left off, with no re-explaining the repo to start over.
In the era of AI web appswe're building the Mac app.
Tell it what to change. Decide what lands.
One cockpit for the AI coding agents you already use, like Codex, Claude Code, and more on the way. Direct the active one by voice or text, and switch without losing your place.
Live in betamacOS
Coming nextWindows
Coming nextLinux
The problem
AI coding tools still lose the plot.
You start with one AI tool. It learns your project for thirty minutes. Then the context gets noisy, the session dies, or the limit hits. Switch providers, and you re-explain the whole repo from scratch.
The fix isn't another chatbot. It's a home for that work that sticks around, instead of scattering across terminals, tabs, and chat logs you'll never reopen. That's the loop Oplyr keeps in one place.
Product
Meet Oplyr.
Your cockpit for AI coding agents.
Oplyr sits between you, your project, and the agents you already use, in a single macOS window where you run them, watch what they're doing, and keep the work on your machine.
“Refactor the auth middleware to use JWT tokens”Listening
- export function checkAuth(req) { return req.session.user; }
+ export async function checkAuthJwt(req) { const token = req.headers.authorization?.split(' ')[1]; }
A local brain, shared across agents
Oplyr remembers your decisions, conventions, and the shape of your codebase as you work, then feeds the right context back to whichever agent is driving, so switching from Codex to Claude Code never means re-explaining the repo. It runs on your Mac, matches on meaning instead of keywords, and tracks which agent taught it what.
Voice or text, never a black box
Say a task or type it. Speech is transcribed right on your Mac and the text stays on screen, so you read and edit what the agent heard before it acts.
Every change arrives as a diff
Edits land as GitHub-style diffs you review before anything touches disk. Read them line by line, keep the good ones, send the rest back, and files change only when you say so.
Fenced to the project you pick
Work stays inside one project root, and agents are steered away from your secrets, with .env files, keys, and credentials on a blocked list. A real terminal and a live codebase map sit in the same window, so the whole loop happens in one safe place.
Built-in terminal
A real shell, right where you work.
Run commands, branches, tests, and provider CLIs without leaving Oplyr. The terminal lives beside your voice session and diff review (same window, same project context) so you stay in one local loop instead of juggling tabs.
Bring your own AI
Use the tools you already trust.
Codex and Claude Code, doing what they do best, in one place. Each runs through your own CLI and login, and moving between them is a click, not a fresh start.
In the beta now
See your codebase.
Don't just vibe through it.
The moment you connect a repo, Oplyr scans it into a live map (every file, folder, and the calls between them) and writes a plain-English summary of each file in the background.
Read it as a tidy tree or spread it out as a force graph, then get a working mental model of your own code back, instead of guessing blind at a repo you half remember.
- Tree and force-directed layouts
- Live file & function dependency graph
- AI file summaries, generated in the background
In the workspace
Markdown, rendered
right where you work.
Open a README or a design note and Oplyr shows it as clean, formatted docs, with no separate viewer and no detour to the browser.
- Headings, lists, tables, and fenced code
- Sits beside the project the doc belongs to
- Read-only by default, nothing edits on open
How it works
One app. Your agents.
Your approval.
Install the macOS app
Oplyr runs locally and walks you through enabling voice input and connecting your first agent.
Connect your agents
Link the provider CLIs you already use, like Codex and Claude Code, and choose which one is active. Add more and switch whenever you like.
Work the loop
Open a project, say or type a task, read the transcript and the diff, and apply only what you want.
Why Oplyr exists
Not another wrapper.
A control layer.
Scattered workflow
Oplyr
A new chat per tool
One cockpit for every agent
Re-explaining the repo
Switch agents, keep the context
Locked to one provider
Bring the agents you already pay for
Pasting code on faith
Read the diff, then approve
Typing everything
Talk or type
Your code on someone's cloud
Runs on your Mac
Where Oplyr fits
Not an IDE. Not an agent that runs off on its own.
Most AI coding tools land in one of two camps. Oplyr is a third thing, a local control layer above the agents you already use, where you stay in the loop and nothing lands until you say so.
The AI IDE
Cursor · GitHub Copilot · Windsurf
Their way
A new editor you move into, with one assistant baked in. You adopt their editor and their provider, and you're living in a chat sidebar all day.
With Oplyr
Keep your editor. Oplyr is a cockpit beside it. Bring Codex and Claude Code, the agents you already pay for, drive by voice, and switch between them mid-task without re-explaining the repo.
The autonomous agent
Devin · Replit Agent · Jules
Their way
Hand a task to a cloud agent and hope. It works on its servers, on your code, and you meet the result after the fact, sometimes somewhere you didn't intend.
With Oplyr
Nothing runs off on its own. Every change arrives as a diff you read and approve, the work is fenced to the one project you picked, and it all runs on your Mac.
Oplyr orchestrates agents like Codex, Claude Code, and Gemini. It doesn't replace them. It's the layer that makes directing them fast, spoken, and safe.
What's next
The bigger bets we're building toward.
The beta already ships a trustworthy coding loop and a local brain that remembers across sessions. From here, Oplyr grows into a durable home for your work: agents that talk to each other, notes that wire back to code, and a vibe layer for the long build. Three bets, equal weight.
Multi-agent room
Agents that talk to each other.
The local brain already lets every agent share what it learns. Next, a room where you bring Codex and Claude Code into the same conversation, let them compare approaches on a hard call, and keep the final say. Conducted by you, never running off on their own.
Notes & meetings
A note-taker that speaks code.
Standups and design reviews captured as you talk, then wired back to the files and decisions they touch, so the important parts don't live only in someone's head.
Vibe layer
A soundtrack for the build.
Focus sprints, debugging spirals, the 2 a.m. shipping runs. Oplyr sets the tempo without pulling you out of the work. We're not saying more yet. 🎧
Directions may shift during beta. We make the core loop trustworthy before expanding.
Trust model
Four promises we won't walk back.
However the product grows, these stay put. Your code, your machine, your call.
Nothing writes without a yes
Agents propose, you approve. No file on disk changes until you say so, and there's no auto-apply switch to forget about.
Speech stays on your Mac
Voice is transcribed on-device, right on your Mac. Your audio is never uploaded to Oplyr or any cloud service.
Work stays in bounds
Each session is fenced to the project root you pick, so an agent can't reach outside it.
Your accounts, your machine
Oplyr never runs your code on our servers and never holds your provider logins. The CLIs talk to their own services under your accounts.
Real talk
Reasons not to bet
on us, yet.
Be skeptical. Here's why.
- 01We're late, and the field is stacked.
The well-funded incumbents already have the users, the budgets, and the head start. We didn't get here first, and switching costs you something.
- 02It's a rough early beta.
Apple Silicon macOS only. Things break, features change week to week, and none of it is ready to carry your production work yet. Use it because it's interesting, not because it's finished.
- 03We're a small, unfunded team.
No enterprise SLA, no big logo behind us, no war chest. If we lose momentum, no one bails us out. That's the real risk of betting on us this early.
- 04We might be betting wrong.
If fully autonomous agents end up just working, a layer built to keep you in control matters less. We don't think it plays out that way, but we could be wrong.
The deal
Free. Local. Yours.
Oplyr is free for developers, with no subscription, no token meter, and no account needed to do the work. We're not chasing revenue yet. We're chasing developers who want the golden era of coding back, where you stay in control of every line the AI writes.
$0, free for developers
No subscription, no per-seat pricing, no tokens billed on top. Use it as much as you want. We'd rather have developers who help shape the product than a paywall between you and it.
Everything stays on your machine
Your code and repos never touch our servers, and speech-to-text runs entirely on your Mac, with no audio upload and no cloud transcription bill. Local by default, private by design.
You decide what ships
Every change arrives as a diff you approve. Oplyr won't merge behind your back or run on autopilot. The AI proposes, and you stay the one who commits.
Bring the agents you already have
One cockpit for multiple providers, and you switch the active one mid-task. Never locked to a single model or vendor. Your accounts, your call.
Early access
Oplyr is opening slowly.
We're inviting developers who lean on AI coding tools daily and want them in one local workspace instead of a dozen scattered tabs.
0downloads and counting
2 fields. No nonsense.
Join the beta list.
Name and email only. No spam, no data games. Our evil empire budget is currently zero. 🫡
Mac-first beta seatsHuman invite reviewOne tiny form
Feedback / questions
Have a thought, bug, roast, or question?
Send it directly. Plain text only, no attachments, no weird payloads, no hacker cosplay.