Perplexity launches self-improving memory system Brain

// 93d agoPRODUCT LAUNCH

Perplexity launches self-improving memory system Brain

Perplexity Brain is a self-improving memory system for the Perplexity Computer agent that builds a living context graph of actions, successes, and failures. Consolidating this graph overnight to optimize instructions reportedly increases task correctness by 25% and reduces execution costs by 13%.

// ANALYSIS

Moving from static prompt engineering to dynamic self-optimizing instructions is a crucial shift for reliable AI agents.

–Shifting focus to an agent's memory of its own execution history enables true continuous improvement.
–Offloading the consolidation of context graphs to overnight execution avoids runtime latency penalties.
–The reported 25% correctness boost demonstrates the practical value of systematic self-reflection loops.

// TAGS

perplexity-brainagentagent-memoryself-improving-agentsperplexity-computer

DISCOVERED

93d ago

2026-06-20

PUBLISHED

93d ago

2026-06-20

RELEVANCE

8/ 10

AUTHOR

DIY Smart Code

// KEEP READING

More AI developer news from the feed

EXPLORE FULL FEED

BENCHMARK1h ago

Jev 1.13 beats Cohere, Voyage in latency

Michael Chomsky shared latency benchmark results comparing TypeSafe AI's Jev 1.13 model against leading dedicated rerankers for search and retrieval pipelines. In the test, Jev 1.13 registered the fastest speeds with a 149 ms median and 205 ms p95 latency, beating Cohere v3.5 (176 ms median / 357 ms p95), Voyage 2.5 Lite (183 ms median / 232 ms p95), and Cohere 4 Fast (183 ms median / 824 ms p95).

MODEL1h ago

Tripo3D P2 lands on fal with decoupled controls

Tripo3D has launched its P2 model on fal, providing fast API-driven text-to-3D and image-to-3D textured mesh generation for creators and developers. The model introduces four texture-quality tiers, decoupled seeds for geometry and texture to facilitate fine-grained iterative design, and support for removing baked-in lighting via the v3.5 texture model for seamless integration into real-time 3D pipelines.

MODEL2h ago

TypeSafe AI has launched Jev, a "System One" AI model designed to make fast, structured decisions for software automation.

Jev is a new frontier model developed by TypeSafe AI that focuses on making structured, probabilistic decisions instead of generating free-form text. Dubbed a "System One" model, it excels at rapid judgments like routing, scoring, and data extraction, delivering typed responses with calibrated confidence levels in just 70 to 500 milliseconds. This approach allows developers to seamlessly integrate AI decisions directly into their code logic with dramatically lower latency and cost than traditional LLMs.