OpenAI

Run GPT-4o, o1, and your fine-tuned models inside AgentAI with full streaming, tool use, and structured output support.

What You Can Do with OpenAI + AgentAI

• Use Any OpenAI Model: Run GPT-4o, GPT-4o-mini, o1, o1-mini, and your custom fine-tuned models across every agent in your workspace.
• Stream Responses in Real Time: Token-by-token streaming for chat-based agents, with full support for parallel tool calls and structured outputs.
• Generate Embeddings at Scale: Power vector search, semantic memory, and RAG workflows with native embeddings support.
• Mix Models per Task: Route reasoning-heavy work to o1, conversational tasks to GPT-4o, and high-volume tasks to GPT-4o-mini, all from the same agent definition.
• Track Token Usage Automatically: Every prompt and completion is logged with token counts, costs, and latencies, so you always know what each agent is spending.

How It Works

1. Generate an API Key. Log into your OpenAI dashboard and create an API key with the permissions you need.
2. Connect to AgentAI. Add the key to your AgentAI workspace as OPENAI_API_KEY in your environment settings. The connection is encrypted end-to-end.
3. Configure Your Agents. Point any agent at any OpenAI model with a single line of configuration. Switch models per task or per agent without rewriting your setup.
4. Run and Monitor. Every call streams through AgentAI's tracing layer. View prompts, responses, token usage, and costs in real time from your dashboard.

Pricing

• Available on all AgentAI plans, including the free Hobby tier
• No markup on OpenAI token costs — you pay OpenAI directly at their published rates
• Unlimited agents, unlimited calls — AgentAI does not meter API usage
• Free to enable from your workspace settings, ready in under five minutes

OpenAI + AgentAI gives every team production-grade access to the most widely-used models in the industry, with the orchestration, governance, and observability layer your agents need to run reliably. Stop writing custom adapters, stop guessing at token costs, and stop worrying about runaway loops. Plug in your key, point your agents at the right model, and ship.

‍