Founding 500: $29 a month, locked for life. 417 machines left.
Tweety Bird
Install

Text Claude Code and Codex from your phone. They share one memory.

Tweety Bird runs on the Mac you already own. Kick off a migration from the taco line, walk away, and get the diff back in the thread. What Claude decided at 11, Codex knows at 2. Your code and credentials never leave your machine.

Install in one commandSee a Wednesday

One command. Card is not charged until it proves it remembers. Reply "goldfish" for a full refund, any time in 60 days.

Tweety Bird, a yellow bird wearing a headset
Tweety Birdbot · your Mac is connected
Wednesday

how's the migration

Claude finished the billing tables at 12:06 and has been waiting on a permission prompt to drop legacy_invoices. 47/47 pass on the new schema.

You told me Thursday to keep Stripe customer IDs denormalized on the org row, so I left that alone. Diff attached.

Approve the drop?

billing-migration.diff3 files · from your Mac

One command. It finds Claude Code and Codex, pairs your phone, and walks you through the proof in under five minutes.

Proves it before it charges you. Tell Claude one decision. Ask Codex. If Codex does not know, you were never charged.

Nothing leaves your Mac. Your repo, your env vars, your cookies, your MCP servers stay where they are. The phone reaches your own box.

Works in Telegram, iMessage, WhatsApp, Slack DM, and Discord. Works with Max, Pro, Plus, API keys, Bedrock, and gateways.

You are the RAM.

You pay $200 a month for two of the best engineers ever built, and you work for them as their memory.

Every session starts the same way. You re-explain the monorepo. You re-state the conventions. You re-identify the bug from yesterday. Auto-compact fires somewhere around 160K and quietly deletes the constraint from message 20. Claude hits the 5-hour window at 3:30, you move the work to Codex, and Codex has never heard of the thing you spent the whole morning on. So you type out the Stripe decision again. Third time this month. Codex flags it as a smell anyway.

Anthropic's own issue tracker has the sentence for this. Issue #39961: "The user ends up serving as the continuity layer between sessions, which defeats the purpose."

Groundhog Day, with the most important tool you own.

Tweety Bird fires you from that job. One thread on your phone reaches both agents on your own Mac. Under both of them, one memory. What Claude decided at 11 AM is what Codex knows at 2 PM. What either of them learned last Thursday is still there Monday. It survives compaction, survives the cap, survives switching machines, survives switching labs.

One thread. One memory. Your machine.

How the memory works →

Six things it does, written the way you would say them.

One thread. Both agents. Still there tomorrow.

Text it like a coworker. It answers from your machine, your repo, your MCP servers, your env vars. Claude and Codex both live in the same thread, so you do not switch apps to switch agents. When it hits a permission prompt at 12:06, you get the prompt on your phone at 12:06, with the diff.

Anthropic gives you a phone window into one Claude session. OpenAI gives you a phone window into Codex. Tweety Bird gives you one thread into both, on your machine, that is still there tomorrow.

Your laptop closes at 6. Your agents don't.

Lid closes. Agent keeps going on the mini. The overnight refactor hits a prompt at 11:22 PM, your phone buzzes at 11:22 PM, you tap, it finishes by 2 AM instead of stalling for seven hours. Read-heavy work runs unattended. Anything that spends money or touches prod waits for your tap.

Cloud agents work while you're away by taking your code somewhere else. Tweety Bird works while you're away on the Mac you already own.

Your agent sends the file back.

"Send me the diff as a file." You get a file, not a summary that says "done." The screenshot of the failing Playwright run, in your thread, in ten seconds, no scp. It knows which file "the report" is, because it is the same agent that just made it.

Every remote lets you send a photo to your agent. Tweety Bird lets your agent send the file back.

Codex already knows what Claude decided.

Ask Codex to review the migration Claude wrote. It opens the PR already knowing you keep Stripe customer IDs on the org row, and why. It does not flag it. Claude hits the 5-hour cap at 3:30; you text "codex, take it from here," and Codex takes it from there, mid-thought.

Anthropic remembers for Claude, on one machine. OpenAI remembers for Codex, if you turn it on. Tweety Bird remembers for both, and they remember each other.

It clicks the button you would have opened the laptop for.

A real Chrome, on your Mac, logged into the dashboards you are already logged into. First of the month: Stripe, AWS, Vercel, Supabase invoices, pulled and texted as PDFs. Every morning: it clicks through the real signup flow on prod and only texts you if something broke.

Claude in Chrome needs you at the desktop. Codex computer use needs the desktop app. Zapier needs an API. Tweety Bird needs a text.

Figure it out once. It runs at 7 AM forever.

Any prompt you have gotten right once becomes a schedule. The scheduled run has today's memory, not a fresh clone that has never met your repo. Actual cron on your box, not a 7-day timer that dies with your terminal. If the Mac was asleep, it catches up.

Every native scheduler either dies with your terminal or starts from a fresh clone. Tweety Bird's runs on your box with today's memory and texts you the result.

A Wednesday, with Tweety Bird.

  1. You kick off the billing migration in Claude Code, close the lid, and walk to lunch.

  2. Claude hits a permission prompt. Your phone buzzes in the taco line: the prompt, the diff, three buttons. You tap "Show me," read it on a screen you can actually read, tap Yes. Fifty-five minutes of agent time you would have lost, kept.

  3. "codex, review what claude did on billing." Codex opens the PR already knowing the Thursday decision. It finds an actual bug in the retry path instead. You did not re-explain anything.

  4. The long refactor hits a prompt. Your phone buzzes once. You read the diff in bed, tap yes, go to sleep. It finishes at 2:04 AM.

  5. The first thing you check is not GitHub. It is a text: "Refactor done, PR #412 open, CI green, 3 files you should eyeball." Attached: the diff as a file.

See the whole day →

Why neither lab will ever ship this.

Anthropic shipped memory. It is machine-local and Claude-only. OpenAI shipped memory. It is off by default, tied to a single codex_home, and Codex-only.

Anthropic is never going to teach Claude to remember what Codex did. OpenAI is never going to teach Codex to take notes for Claude. The most valuable memory in your entire workflow is the one that crosses the vendor line, and there is a structural reason nobody at either company is positioned to build it. Vendor lock-in, dressed as memory.

You run two labs because two labs beats one. You should not have to be the cable between them.

The cross-tool story →

The stack you would need to match it.

Remote Control for Claude. The ChatGPT app for Codex. A memory MCP wired into each agent separately. A Telegram bridge that runs with --dangerously-skip-permissions because that is the only way it works unattended. A scheduler that either expires in seven days or starts from a fresh clone. A browser extension that only works when you are at the desk.

Five installs. Four vendors. Two research previews. Zero shared state.

Tweety Bird is one command, and nothing in it lives inside either agent. They update; nothing breaks.

See the whole table →

If you are a one-person company.

You do not have an engineering department. You have Claude Code, and Claude Code is the reason you can build software at all. Tweety Bird gives it hours you do not have and hands you do not have. First of the month, it logs into Stripe and AWS and pulls the invoices while you sleep. Every morning it QAs your signup flow in a real browser and only texts you if something broke. Friday at 5 it drafts the build-in-public post from what actually shipped, you tap approve, it posts. No DevOps afternoon. No VPS.

If you are not a developer at all.

If you run Claude Code across two machines for a research pipeline, a curation workflow, a model you maintain, you already know the bug: the agent re-implements a fix it committed yesterday and asks "didn't I just say this?" Anthropic's auto memory is machine-local by design. Tweety Bird's is not. Same memory on the NUC and the ThinkPad, same memory next week, and a way to check on a six-hour run from your phone without SSH. A research assistant that stops getting amnesia.

I built this because I was the RAM.

I run two labs. Claude Code for the ambiguous architecture work, Codex for the well-scoped stuff and the review pass. I have a MacBook that closes at 6 and a Mac mini that does not. I had tmux, Tailscale, Termius, a Telegram bot I wrote in a weekend and then maintained for six months against my will. I re-explained the same Stripe decision to Codex an hour after Claude and I made it. I was paying $300 a month to be the continuity layer for two agents that were each smarter than me at the thing I was doing for them.

So I built the harness I wanted and lived in it. For most of a year, my agents have answered my phone. Tweety Bird is that harness, packaged into one command, with the sharp edges filed off.

John Rood

The whole story, and why 500 →

Nobody screenshots a note pile.

"It's a goldfish."
anthropics/claude-code issue #14227
"The user ends up serving as the continuity layer between sessions, which defeats the purpose of an AI coding assistant."
anthropics/claude-code issue #39961
"Before compaction, Codex reported progress around 97%. After context was automatically compacted, Codex reported progress around 42%."
openai/codex issue #25792
"My biggest issue with Codex is not code generation, but context loss."
OpenAI Developer Community
"I'll kick off a task, walk away, come back and the connection dropped with no way to recover the session."
Hacker News, on Remote Control
"It was like having Groundhog Day with my most important development tool."
The Production Line

These are the people Tweety Bird is for. Some of them are already in the founders' thread.

Coffee-a-day money for a night shift.

Founding 500

$29 a month, locked for life. Or $290 a year. First 500 machines.

Why 500 →

Solo.

$39 a month.

Or $348 a year. Founding 500: $29 a month, locked for life.

Two Macs. Both agents. One thread. One memory. Permission relay. Files back. Browser hands. Schedules. CLAUDE.md ingestion. The Night Shift pack. The founders' thread.

Install in one command

Team.

$149 a month for five.

$25 a seat after.

Everything in Solo, and your team's agents share one brain per repo. A new hire's agent knows every decision on day one. Admin view of what ran unattended and who approved it.

Start with five

Card is not charged until Codex recalls a decision you gave Claude in your first session. 60 days, reply "goldfish" to any email for a full refund. Cancel from the thread with one message.

You pay $200 a month for agents that work about nine hours a day, because that is how long the lid is open. $29 buys the other fifteen.

Everything on the plans →

The eight you were about to ask.

Does my code leave my machine?

No. Claude Code and Codex run on your Mac, the way they do now. Your repo, your env vars, your cookies, your MCP servers, your uncommitted state stay where they are. The phone reaches your own box. This is the same architecture Anthropic chose when they made Remote Control outbound-only and OpenAI chose when they made the Mac the Codex worker.

Is this another note pile the agent never reads?

The install makes you find out in ten minutes. Tell Claude one real decision. Open Codex. Ask it. If Codex does not know, you were never charged. The memory is written and read on both agents' lifecycle hooks, every turn. It is not a file the agent has to be told to check. It is not off by default.

Does it survive compaction?

Yes, because it was never in the context window. Auto-compact summarizes what is in the window. The "why," the constraint from message 20, the rejected approach, all live in memory and get recalled when they are relevant again. 97 percent stays 97 percent.

Does it run --dangerously-skip-permissions?

No. Tweety Bird relays permission prompts to your phone with the diff, and you set the policy per repo: what runs unattended (reads, tests, lint, PRs to a branch) and what waits for a tap (spending money, pushing to prod, messaging anyone, deleting anything).

Does it work with my plan?

Any plan. Max 5x, Max 20x, Pro, Plus, ChatGPT Pro, API keys, Bedrock, a gateway, a custom base URL. Tweety Bird runs beside your agents and uses the logins you already have.

I only use Claude Code. Or only Codex. Is it still worth it?

Yes. One agent with a phone thread, memory that survives compaction and machines, files back, browser hands, and schedules with today's memory is already the product. The second agent is free when you want it.

Does the Mac have to stay awake?

The agent has to be running somewhere. On the mini, that is a box that never sleeps. On the MacBook, if the lid is closed and it sleeps, work pauses and resumes when it wakes, with catch-up for anything scheduled. It does not need a terminal window open or a desktop app in the foreground.

What if it forgets?

Reply to any email with the word "goldfish." Full refund, 60 days, no form, no call. We also want the screenshot, because that is a bug and founders' bugs ship first.

Read every objection →

Stop being the RAM.

One command. Five minutes. Tell Claude one decision, ask Codex, and watch it answer. Your card is not touched until it does.

Then close the lid at 6 and see what you woke up to.

Install in one command

$29 a month, locked for life for the first 500 machines. 417 left. Reply "goldfish" for a full refund, any time in 60 days.

Your laptop closes at 6. Your agents don't.