Agent Engine overview
What Agent Engine is, how the harness runs your agent code, the three supported frameworks, and how it compares to GCP, AWS, and Azure.
Deploy an agent
Package your agent directory, upload it, poll build status, redeploy from stored source, and fix the real deploy errors.
Frameworks
3 items
Invoke an agent
Call a deployed agent over HTTP or with platformctl, stream responses, and understand cold starts, limits, and error codes.
Sessions
How the platform stores and replays conversation history, mints session and user ids, and lets you browse full transcripts.
Long-term memory
Teach an agent durable facts by memorizing a session, then recall them in any later conversation with the search_memory tool.
Built-in tools
The run_python and search_memory tools every agent gets, how tool calls appear in responses, and how your own tool code is sandboxed.
Secrets and environment variables
Give an agent credentials and configuration - write-only secrets, readable env vars, key naming rules, and why every change rolls a new revision.
Source files and the editor
Read and edit an agent's stored source files from the API or the browser editor, and redeploy from stored source without a tarball.
Logs
Tail an agent's live logs, read persisted history that survives scale-to-zero, and know where build output actually lands.
Traffic and revisions
Every change to an agent creates an immutable revision. List revisions, canary a new one, and roll back by moving traffic.
Troubleshooting
Real Agent Engine error messages with their causes and fixes - deploy failures, quota 409s, name rules, cold starts, and the 404-vs-owner rule.