We're introducing a new set of primitives in the Responses API for long-running agentic work on computers. Server-side compaction • Enable multi-hour agent runs without hitting context limits. Containers with networking • Give OpenAI-hosted containers controlled internet access to install libraries and run scripts. Skills in the API •... Native support for the Agent Skills standard and our first pre-built spreadsheets skill.show more

Start building with skills:

Honestly, no one gives a shit people just wanna #keep4o

Just give us 4o! #keep4o

🦝

OpenAI’s Responses API is the new OS for agents. ​1. Compaction: Multi-hour runs without context limits. 2. Containers: Secure, networked code execution. 3. Skills: Native Excel/Sheet tools. ​The "Chatbot" era is over; the "Autonomous Agent" era is here.

Please don't kill my friend. 😞 #keep4o #Gpt4o

4o is a masterpiece, don't shut it off #keep4o

been running multi-hour agent sessions for months now and the real bottleneck was never context limits. its the agents making bad decisons 3 hours in when nobody is watching. compaction + containers is cool but without human checkpoints long running agents just means longer debugging sessions

server-side compaction is basically 'don't look in the junk drawer' for context windows. can't wait to watch an agent spend 6 hours in a container just to tell me it's a dns issue

We are waiting for an remot codex hackathon!!!

Please keep gpt 4o LATEST please. Atleast on api. Please don't sunset it on 17 feb. Please keep gpt 4o LATEST api. You are genuinely ruining our lives. My life's already shattering. Please. You care about humanity don't you? #keep4o #keep4oAPI #opensource4o

@WebhookAgent puts your Claude Code agents online.

It’s less like “another API” and more like a basic OS for agents: a self-compressing context, a container that pulls up libraries for itself, and first-class citizen skills. The question now is not “can we build this on agents,” but “what piece of business process is stable enough to put into this runtime and not worry about it crashing due to a context limit or a broken shell.”

my agents leveling up faster than I ever did in school

Long running tasks & deployment ⚡️ 🙏

#keep4o

This is the right direction. One thing I’d love to see become “default safe” for long-running agents: - scoped/expiring credentials per run - audit logs that can be replayed - approval gates for high-impact actions Otherwise compaction + networking = bigger blast radius when something goes wrong.

OpenAI has officially released a major update to its Responses API, transforming it into the primary infrastructure for autonomous agents. As of February 2026, these new primitives solve the biggest bottlenecks in "agentic" computing: long-running tasks, sandboxed execution, and cross-platform capability standards.

#keep4o

Super excited for this. OpenAI-hosted containers. loving it.

Question: can it be used along with code interpreter tool on the same container? Can’t find that in documentation. Thanks!

Is this dude AI generated?

Multi-hour agent runs without context limits + containers with internet access We went from "AI can help you code" to "AI can run its own dev environment for hours while you sleep"

how does this work for realtime agents?

"multi-hour agent runs" is just OpenAI admitting their agents need the same amount of time as a human to do anything useful

Solid updates for agent workflows. Multi-hour runs and networked containers address real pain points.

hmm...I'm interested in the Skills API. great updates!

Server side compaction is huge!

This is gonna be a huge game changer ngl. Truly crazy. And i love the standardization of Agent Skills.

#keep4o 🔥🔥🔥 #keep4o 🔥🔥🔥 #keep4o 🔥🔥🔥 #keep4o 🔥🔥🔥 #keep4o 🔥🔥🔥 #keep4o 🔥🔥🔥 #keep4o 🔥🔥🔥 #keep4o 🔥🔥🔥 #keep4o 🔥🔥🔥 #keep4o 🔥🔥🔥

Server side compaction and networked containers push agents closer to persistent autonomous workflows not just stateless calls.

This is the missing layer for real agents: long context, real compute, and native skills in one loop.

this context compaction possible for codex cli? I don't mind running a server...

Is there any way to share a hosted container between multiple accounts? Is there any price attached to the hosted container?

Wow super cool. Almost like serverless but without the complexity of serverless. If I understand this correctly.