Skip to main content
No. Your agent calls the model provider you choose, with its API key added as a secret. Onecortex runs your agent, not a model.
Python 3.12 and 3.11, and TypeScript or JavaScript on Node.js 22 and 20. LangGraph, the OpenAI Agents SDK, Strands Agents, CrewAI, LangChain and LlamaIndex in Python; Mastra, the Vercel AI SDK and LangChain.js in TypeScript; and any agent at all as a plain function. See Entrypoints.
No, by design. Onecortex builds the image from your dependency file so that every agent gets the same tested runtime, and so that a build can be smoke tested the same way every time. systemPackages and build.commands cover most needs. See Python agents.
No. Point entrypoint at your agent object, and Onecortex serves it. A server started at import time fails the build. See Entrypoints.
Never. The GitHub app has read only access, and your source is never modified. See Security.
No. Onecortex hosts agents. Your agent can call MCP servers, and calling your agent from MCP clients is coming: see MCP.
Not yet. Your agent’s logs and the event stream, with every tool call and step, are how you see what a run did.
Nothing, during the private beta. See The private beta.
In Europe (Frankfurt). See Regions.
Nothing. It keeps serving the version that was live. A new version goes live only after it has passed its smoke test. See Builds.
Only what it keeps itself. Calls with the same sessionId reach the same running instance, so in memory state survives between them for a while. See Sessions.
Not yet. Every agent is called at https://api.onecortex.io.
Last modified on September 28, 2026