OpenResponses vs Pounce

OpenResponses is OpenAI's API on your iPhone and iPad, with nothing hidden. Bring your own API key and it shows you the request before it goes out, the events as they stream back, and every tool call along the way. It's for people who build with OpenAI's models, or who just want to see what the newest ones actually do before writing any code. It started as the successor to OpenAssistant, my app for OpenAI's older Assistants API, and talks straight to OpenAI's Responses API with no server of mine in between. PICK A MODEL AND SEE WHAT IT DOES • GPT-6.1 Sol, GPT-6 Sol, Astra and Luna, the GPT-5.6 family, and older models your key can still use. When OpenAI releases a new one, it shows up in the menu on its own, with its settings read from OpenAI's docs. • Reasoning from None to Max where the model takes it, Pro mode, verbosity, context compaction and the rest of the request settings, all in one place. • Replies stream in as they're written, with the reasoning summaries the API sends, token use, and the full request and response a tap away. • Save setups as presets, attach photos and documents, export conversations, and search your chats by what they're about, not just the exact words. THE TOOLS THE API HAS • Web search with sources, Code Interpreter, file search over your own vector stores, and image generation and editing with GPT Image 2.5, up to Max quality. • Voice conversations with GPT Realtime 2.1 or GPT-Live 1, with mute, and with headphones and Voice Barge-In on you can talk over it to cut it off. • A browser the model can drive, inside the app, with Stop, time limits and action limits. • MCP connections. Sign in to supported apps or add a server by address, look through the tools it offers, and approve tool calls unless you turn that off. • Calendar, Reminders, Contacts and Notion, only when you turn them on. • Local Python. The model writes it, you read it, you tap Run, and it runs on your phone. Standard library, no internet, 30 seconds a run. API WORKBENCH For when you'd rather write the request yourself. Edit the raw JSON, send it over HTTP, streaming or WebSocket, watch every event come back, and export the whole response. It also counts tokens, compacts context, and manages conversations, files, vector stores, containers and batch jobs. Anything that deletes asks first. A few things worth knowing before you download it: • There's no OpenResponses account. You need your own OpenAI API key, and what you use is billed to your OpenAI account. • Your key stays in the iOS Keychain and only ever goes to OpenAI. • Explore Demo lets you look around without a key, and it sends nothing. • Before your first live request, the app shows what gets sent to OpenAI and waits for you to say yes. • It runs on iPhone and iPad with iOS 17 or later. One last thing. This shows you what OpenAI's models do, and they can be wrong, sometimes confidently. Check anything that matters before you act on it.

Pounce puts your AI coding agents in your pocket. Start Pounce on your computer, scan the pairing code, and every agent session running there appears on your phone. Watch them think, write code, run commands, and open pull requests in real time — then step in whenever you want. WHAT YOU CAN DO • Follow every session live — messages, reasoning, tool calls, terminal output and git activity in one clean timeline • Review the diff, then commit, push, or open a pull request • Send a new instruction or answer a prompt with a tap • Use your voice — hold and speak to steer an agent hands-free • See where your usage goes — tokens, spend, sessions and messages, broken down by agent, by model and by project • Keep every machine and project in one list, sorted by what needs you WORKS WITH Claude Code, Codex, Cursor and opencode. HOW IT WORKS 1. Start Pounce on your Mac, Windows or Linux machine — the desktop app, or a single command in any terminal 2. Scan the pairing code with Pounce 3. Your agents and projects appear automatically REACH IT FROM ANYWHERE Pounce connects straight to your own computer over an encrypted peer-to-peer link. That works on cellular or a café network with no VPN and no port forwarding, and your code and conversations never pass through a server of ours. Pounce is open source under the MIT license. Whether you're grabbing coffee, on the couch, or away from your laptop, Pounce keeps your agents moving.