Perplexity Brain: AI Agent Memory That Learns From Its Own Work Overnight

Perplexity’s new Brain gives its AI agent a memory of its own work, not just your preferences. It reviews each day’s tasks overnight and, in early tests, gets 25% more answers right on jobs it has done before while cutting their cost by 13%.
artificial-intelligence
Author

Kabui, Charles

Published

2026-06-23

Keywords

perplexity-brain, agent-memory, self-improving-ai, context-graph, ai-agents

Download as PDF

Perplexity launched Brain on June 18, 2026, a memory system for its AI agent, Computer, that learns from its own work. Most AI memory stores facts about you: your preferences, your role, how you work. Brain does the opposite. It records what the agent did, which steps worked, which failed, and what you corrected, then keeps these notes as a context graph: a linked, plain-language wiki of your projects, people, and sources. Overnight, it rereads that map and rewrites it to work better next time, and every entry links back to its original source. In Perplexity’s early tests, Brain raised answer correctness by 25% on jobs it had done before, lifted recall by 16%, and cut the cost of work needing past context by 13%.

This turns memory from a fixed profile of you into a skill that builds up with use. Take a developer fixing bugs across several repositories: Brain remembers which files mattered last time, so the agent stops relearning that context and reaches the cause with fewer model calls. For teams who pay by the token, the text these models bill for, today’s work makes tomorrow’s cheaper.

An agent that rewrites its own playbook overnight behaves less like a tool you reconfigure and more like a worker who improves with experience. The limit stops being how much fits in the context window and becomes how well an agent digests what it already learned. These numbers are early and first-party.

Read More: OpenAI’s Dreaming V3 builds the opposite kind of memory, a profile of you that it also refreshes overnight.

Sources:

Disclaimer: For information only. Accuracy or completeness not guaranteed. Illegal use prohibited. Not professional advice or solicitation. Read more: /terms-of-service

Reuse

GNU GENERAL PUBLIC LICENSE v3.0(View License)

Citation

BibTeX citation:
@misc{kabui2026,
  author = {{Kabui, Charles}},
  title = {Perplexity {Brain:} {AI} {Agent} {Memory} {That} {Learns}
    {From} {Its} {Own} {Work} {Overnight}},
  date = {2026-06-23},
  url = {https://toknow.ai/posts/perplexity-brain-self-improving-agent-memory/},
  langid = {en-GB}
}
For attribution, please cite this work as:
Kabui, Charles. 2026. “Perplexity Brain: AI Agent Memory That Learns From Its Own Work Overnight.” https://toknow.ai/posts/perplexity-brain-self-improving-agent-memory/.