L3 Context and memory
External memory
Keeps what the agent must remember outside its context window and brings back only what matters for the request.
Why it matters
The context window, everything the model sees in one call, starts empty each session and loses detail when compacted, so without memory the agent forgets preferences and decisions. Done carelessly, memory floods the window, recalls stale facts, or replays planted text as an order.
What to build
Re-read capped instruction files each turn
The default the research recommends. The reasoning and the code behind it are in the app.
What it takes, by kind of agent
- Conversational assistantMust-have
- CodingOptional
- Deep researchMust-have
- Browser/GUINot applicable
How well it is studied
- Harnesses checked
- 30
- Checked facts behind it
- 427
See the 427 checked facts and how each harness does it
The facts, the code lines they come from and the rules for this topic are in the app. Sign in with Google; new accounts are approved by hand.
Also in L3 Context and memory
- CompactionSettled practice
Shrinks the conversation before it overflows the model's context window, without losing what the agent must remember.
- Assistant Must-have
- Coding Must-have
- Research Must-have
- Browser Must-have
- Result clearingSettled practice
Keeps large tool outputs out of the conversation while leaving the full text one request away.
- Assistant Must-have
- Coding Must-have
- Research Must-have
- Browser Must-have
- Working planOpen problem
Keeps the agent's current plan, such as its to-do list, in its own saved place and shows it to the model on every turn.
- Assistant Must-have
- Coding Must-have
- Research Must-have
- Browser Must-have
- Retrieval qualityOpen problem
Finds the few passages that answer a question and gives each one an id the answer can cite.
- Assistant Must-have
- Coding Not applicable
- Research Must-have
- Browser Not applicable