For OpenAI, from 🤫 hussh
Frontier intelligence, on the person's terms.
ChatGPT showed the world what AI can do. Then gpt-oss did something just as important: it put OpenAI's open weights on the person's own hardware. 🤫 Private Agent One is model-agnostic and consent-first. A person can use GPT by consent while their private data stays on their device, connect over MCP, or run gpt-oss entirely on a machine they own. The same intelligence, reaching people on their terms.
OpenAI intelligence, in the tier the cloud cannot reach.
gpt-oss is your Apache 2.0 open-weight family. The 20B model runs on-device in about 16GB.
🤫 Agent One runs gpt-oss entirely on the person's own hardware, with no cloud round-trip.
OpenAI intelligence in the private, on-device tier that a cloud API cannot serve.
Your agent stack, from ChatGPT agent to AgentKit and the Agents SDK, speaks MCP natively.
Agent One speaks MCP too. A person's agent and a GPT-based agent work together by consent.
Agent-to-agent work on standards you already ship, with a receipt for every consented call.
Regulated buyers want frontier AI and hesitate to send their data to a cloud API.
Agent One is the private client layer: data on the device, a PCHP receipt for every access.
GPT reaches insurance, finance, healthcare and government, where a raw API cannot go.
The frontier, the brand, and open weights.
OpenAI is an AI research and deployment company, now the OpenAI Group PBC under the OpenAI Foundation. It makes ChatGPT, the GPT family and Sora, and runs the platform millions of developers build on.
Announced at DevDay in October 2025. The largest consumer AI footprint in the world, and the name people know.
Building on the API at 8B tokens a minute. The default place to build with AI.
In OpenAI's words, "the fastest-growing business platform in history."
gpt-oss-120b and gpt-oss-20b, released August 2025. The 20B runs on-device in about 16GB.
First-class in the Agents SDK and the Responses API. The same rail 🤫 speaks.
Stargate, AMD (6 GW), Broadcom, NVIDIA and Oracle, taken together. A multi-year buildout across many vendors.
Scale, the frontier, and a real open-weight family that runs on the person's device. That last one is where our worlds meet.
ChatGPT at scale
800M+ people a week. The brand that defined the category and the front door to AI for most people.
The model frontier
The GPT-5 family, reasoning models and Sora. Frontier capability across text, reasoning and video.
The developer platform
The API, the Responses API, AgentKit and the Agents SDK. Where 4M developers build.
gpt-oss
Apache 2.0 open-weight models that run on-device. OpenAI's own technology, running privately on hardware the person owns.
Agents
ChatGPT agent with computer use, AgentKit and the Agents SDK, all with first-class MCP.
Compute at nation scale
Stargate and commitments across many vendors. The infrastructure behind "abundant intelligence."
Where OpenAI is going.
AGI that benefits all of humanity
The mission: abundant intelligence, delivered safely and broadly.
Compute abundance
Stargate, AMD, Broadcom, NVIDIA and Oracle. Compute as the path to progress.
Agents
ChatGPT agent, AgentKit and the Agents SDK, with developers as the distribution layer.
Consumer and enterprise
Growing ChatGPT and business seats together, as two engines.
Hardware
A new device designed with Jony Ive's io team: screenless and aware of context.
GPT-5, gpt-oss, ChatGPT agent and AgentKit.
The newest from OpenAI, and the pieces we would build on first.
The GPT-5 family
Launched August 2025, with GPT-5 Pro and GPT-5-Codex shown at DevDay. The frontier a person can reach by consent.
gpt-oss
gpt-oss-120b and gpt-oss-20b under Apache 2.0, OpenAI's first open-weight models since GPT-2. The 20B runs locally on a phone or PC.
ChatGPT agent and AgentKit
An agent that browses, uses a computer and runs a terminal, plus AgentKit's Agent Builder, ChatKit and connectors. MCP-native throughout.
Responses API and MCP
Remote MCP server support since June 2025. The exact seam where a person's agent and an OpenAI agent meet.
Buy together. Build together. Sell together. Take risk together.
OpenAI brings frontier intelligence at scale. 🤫 brings a private agent on hardware the person owns. Each makes the other go further, and the rails to connect them already exist.
Build together: gpt-oss on the person's device
gpt-oss runs locally under Apache 2.0. The 20B fits in about 16GB.
🤫 runs OpenAI models entirely on the person's hardware. No cloud round-trip, and private data stays home.
The private, on-device tier of OpenAI's mission: intelligence that never leaves the person's machine.
Build together: GPT by consent, over MCP
MCP is first-class in the Agents SDK and the Responses API.
Agent One calls GPT with the person's explicit consent and a PCHP receipt, and offers consent-gated MCP tools that GPT agents can use.
Consented, high-intent GPT usage, and agents that work together on shipped standards.
Sell together: a private client layer for regulated markets
OpenAI wants deeper enterprise reach, and regulated buyers need governance.
Agent One brings consent receipts, on-device data and least-privilege access, built to the FedRAMP High bar.
GPT reaches insurance, finance, healthcare and government. New market for OpenAI.
Take risk together: the edge
OpenAI believes in compute abundance. We believe in the distributed edge.
We tune gpt-oss together for Puppy One and the garage grid, with MCP and A2A for consented agent-to-agent commerce.
Abundance and the edge meet on a node the person owns.
The people building OpenAI.
Sam Altman
CEO (co-founder)
Leads the AGI mission and the bet on compute abundance, to benefit all of humanity.
Brad Lightcap
COO
Runs operations, business and go-to-market, scaling the platform and the enterprise.
Greg Brockman
President (co-founder)
Leads the compute and training infrastructure buildout behind the frontier models.
Mark Chen
Chief Research Officer
Leads the research frontier: the models behind the products.
The same intelligence, on the person's terms.
OpenAI builds intelligence at a scale no one else has reached. 🤫 puts a human-controlled agent on hardware the person owns. Together, GPT reaches people on their terms, on their devices, by their consent, and OpenAI's mission reaches the private tier it was always meant for. Own your AI, own your data, own your compute. And 🤫 Agent One is free to every American.