Every objection, answered.

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. You noticed. So did we.

Decisions and the why behind them. "Keep Stripe customer IDs denormalized on the org row because the join was killing the dashboard." Rejected approaches. Conventions. Promises the agent made. Not your source. Memory entries are encrypted at rest and keyed to you. One command exports all of it as markdown. One command deletes all of it. Team plan can pin memory to your own storage.

No. That flag turns every text into unrestricted code execution and secret exfiltration. 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). Unattended still means you are one tap away.

You. The thread is bound to your account and your paired devices. Nobody who finds your bot's handle can say "yes" to a shell command. Approvals require a paired device, not just a message.

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 the machine sleeps, work pauses and resumes when it wakes, with catch-up for anything scheduled. Tweety Bird is the process that stays alive; it does not need a terminal window open, a desktop app in the foreground, or a session flag you forgot to pass.

The agent on your Mac never noticed. The thread catches up when you do. Nothing dies at ten minutes.

No. Nothing in Tweety Bird lives inside either agent. The memory, the schedules, and the thread sit beside them. When Codex ships a new CLI, your 7 AM audit still fires and your thread still answers.

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, deterministically. It is not a file the agent has to be told to check. It is not a 5,000-token summary extracted after the session goes idle. It is not off by default.

The opposite. Re-explaining is where your tokens go: 94.8 percent of heavy-user tokens are cache reads, roughly 406 tokens read for every one written. Every decision Tweety Bird carries forward is 50 files you do not re-read at 9:30 AM. Recall is small and targeted, not a dump of everything it knows into every prompt.

Yes, because it was never in the context window. Auto-compact fires at 160K and 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.

Keep it or shrink it. Tweety Bird reads it on install and both agents carry what is in it. Most people cut theirs to under 100 lines within a week, because the conventions live in memory now instead of a file that gets ignored past line 200.

Reply to any email with the word "goldfish." Full refund, 60 days, no form, no call. If it forgot something it should have remembered, we also want the screenshot, because that is a bug and founders' bugs ship first.

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. Most people who install for one end up running both, because the 5-hour cap stops being a wall once the other agent can pick up mid-thought.

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. Remote Control refuses API keys and gateways. Tweety Bird does not care how you pay Anthropic or OpenAI.

Telegram, iMessage, WhatsApp, Slack DM, Discord. Whichever one is already open on your phone. You can use more than one.

One command. Under five minutes. It finds your Claude Code and Codex installs, pairs your phone with a code, ingests your CLAUDE.md and AGENTS.md, and walks you through the first-session proof. If it took an afternoon you would already have an afternoon-long setup, and it would be yours, and you would hate maintaining it.

Mac first, because that is where the mini on the shelf lives. Linux is next, for the NUC-and-ThinkPad crowd. Windows after that. Founders vote on the order.

Yes, Solo includes two. Same memory on both. "How's the migration" gets the same answer no matter which box is doing the work.

Team plan gives five people's agents one shared brain per repo. When one person's Codex decides something about the billing service, everyone's Claude knows it. New hires start with the whole history. You see what ran unattended and who tapped yes.

Those are two windows into two sessions in two apps that do not know each other exist. Remote Control exits when the process dies or the network gaps for ten minutes, offers no Auto mode from the phone, and cannot send you a file. Codex mobile needs the Mac awake with the app open and cannot touch the filesystem. Neither remembers. Neither schedules. Neither acts on the web from your phone. Tweety Bird is one thread into both, on your machine, with one memory under them.

OpenClaw is free software and it can do a lot of this if you have an afternoon, a VPS, and an ops habit. It had a critical RCE in February. The guidance is to never expose the gateway and patch continuously. You have a full-time job. Tweety Bird is the packaged version with the sharp edges filed off, a memory that spans both agents by default, and someone whose job it is to keep it safe.

Those are memory stores you wire into each tool yourself, hosted in someone else's cloud, with retrieval limits on the thing whose whole job is retrieval. No phone. No schedule. No hands. Tweety Bird's memory is one part of one install, written and read automatically on both agents, with a phone thread and a scheduler that use it.

Because for the first month, every founding install's memory quality gets reviewed by hand: what got remembered, what got recalled, what missed. That is how recall gets tuned on real repos. We can do that for 500 machines. We cannot do it for 5,000. When machine 500 pairs, the founding price closes.