Self-improving Memory for Agents - Perplexity

2026-06-17 · Source: perplexity.ai via Google News · Field: Technology & Digital — Artificial Intelligence & Machine Learning · Depth: Intermediate, short

Summary

Perplexity has launched "Brain," a self-improving memory system designed for its "Computer" AI agent, available to Max and Enterprise Max subscribers in Research Preview as of June 18, 2026. Unlike traditional AI memory focused on user preferences, Brain builds a context graph of the agent's work, learning from successes, failures, and corrections to enhance performance. This system reviews the graph at set intervals, such as overnight, to teach itself how to execute tasks more efficiently. Early measurements indicate Brain increases answer correctness by 25% and recall by 16% on previously encountered tasks, while reducing the cost of tasks requiring historical context by 13%. Brain achieves this through a recursive self-improvement loop, forming a "living context graph" and an "LLM wiki" that updates incrementally, enabling Computer to become a more proactive AI.

Key takeaway

For AI Engineers or Directors of AI/ML evaluating agent platforms, Perplexity's Brain offers a compelling model for continuous agent improvement. If your team relies on AI agents for complex tasks, you should consider how a self-improving memory system like Brain can significantly boost correctness by 25% and recall by 16%, while cutting costs by 13% for recurring tasks. This investment in agent-centric learning promises more efficient token usage and increasingly proactive AI capabilities over time.

Key insights

Brain enables AI agents to self-improve by learning from their past actions, successes, and failures, enhancing future task performance.

Principles

• Agent memory should focus on its own work, not just user data.
• Continuous learning from agent actions improves task performance.
• Recursive self-improvement is key for proactive AI capabilities.

Method

Brain builds a context graph of agent work, reviewing it overnight to synthesize sessions, connector results, and corrections. This updates an LLM wiki, enabling agents to learn from past actions and improve task execution.

In practice

• Reduce token usage through agent's learned efficiency.
• Agents improve context understanding with continuous use.
• Trace memory entries to original sessions or sources.

Topics

• AI Agents
• Self-improving AI
• Context Graphs
• LLM Memory
• Perplexity Computer
• Recursive Learning

Best for: AI Architect, AI Product Manager, Entrepreneur, AI Engineer, Machine Learning Engineer, Director of AI/ML

Related on AIssential

• How AI Agents Reshape Knowledge Work - Perplexity — AI agents like Perplexity Computer autonomously execute complex workflows, significantly reducing human effort and cost while expanding the scope and cognitive complexity of tasks.
• Your AI Agent Has a Memory Problem — AI agents require robust memory systems to retain context and improve performance across sessions.
• GPT-Realtime-2, Directionally Bad and Agent Memory — Agent memory and context engineering are crucial for enhancing AI agent performance and reducing operational costs.
• 🥇Top AI Papers of the Week — AI advancements focus on novel computing, efficient memory, robust agents, and specialized models.
• Engineering Memory for AI Agents: A Practical Beginner’s Guide — AI agent memory selectively retains past interaction details to enable contextual, multi-step task execution.
• From raw interaction to reusable knowledge: Rethinking memory for AI agents — Structuring AI agent memory into reusable knowledge units improves performance and efficiency across diverse tasks.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by perplexity.ai via Google News.