Long-Running Context
Coding agents and research agents outlive a single HTTP request.
A bookmark and a sticky note, not rereading the novel every morning.
Visual Workflows
Start here — scroll inside each diagram frame to explore, then use + / − to zoom up to 200% if needed.
Overview
Scroll inside the frame to explore · use + / − to zoom up to 200%
Scroll inside the frame to explore · use + / − to zoom up to 200%
This turn
Scroll inside the frame to explore · use + / − to zoom up to 200%
A bookmark and a sticky note, not rereading the novel every morning.
Key Takeaways
- 1.Jobs that last hours need checkpoints of context, not one giant prompt.
- 2.Each resume loads compacted state plus the next slice of work.
- 3.Durable runtimes store that state; you still design what is in it.
- 4.Do not replay the entire history on every resume.
- 5.Coding agents and research agents outlive a single HTTP request.
Learn elsewhere
- →Durable Execution — Phase 21
Real Example
Scenario
Long jobs: pack only what this turn needs.
What you would do
In Context Engineering, apply Long-Running Context to this scenario: Long jobs: pack only what this turn needs. Identify the inputs, run the technique, validate the output, and note one thing you would monitor in production.
Commands
Commands to Remember
Jobs that last hours need checkpoints of context, not one giant promptEach resume loads compacted state plus the next slice of workDurable runtimes store that state; you still design what is in itDo not replay the entire history on every resume
Cheat Sheet
Quick recap
quick ref- •Long jobs
- •Budget
- •Don't dump
Common Mistakes
- ✕Skipping evaluation for Long-Running Context before production
- ✕No logging or tracing around long running context steps
- ✕Ignoring cost and latency implications