[DigitalToday reporter Jinju Hong (홍진주)] Perplexity has introduced a memory system called Brain for its AI agent, Computer, to build up past work history and user edits.

Decrypt, a blockchain media outlet, reported on June 18 that the feature increased answer accuracy for repetitive tasks by 25 percent, improved recall by 16 percent and reduced costs for work requiring extensive context by 13 percent.

Brain focuses on remembering the agent's actual work rather than personal information such as a user's name or preferences. Each time Computer finishes a task, it records in a context graph which connectors it used, which sources were valid, what the user changed and which attempts failed. It then typically aggregates the data overnight to update a personal large language model wiki and loads it into the agent's execution environment before the next task begins.

Perplexity explained Brain by saying, "Instead of starting from scratch every time, Computer starts work with the full context of projects, decisions and sources." It also said, "Each memory is linked to a session, file and source, so you can transparently see what is stored and control it."

The feature is currently offered as a research preview for Max and Enterprise Max subscribers. Memories saved in Brain can be viewed under 'Customize' in the sidebar. The price is $200 a month for Max.

The performance figures presented by Perplexity are its own early metrics, not external benchmarks. The company said improvements were pronounced on tasks it had already handled before. It also made clear that Brain does not make the underlying model itself smarter. It said broad generalisation, such as applying what was learned in financial research directly to coding tasks, remains an unresolved area.

The market views the feature as a case of shifting the focus of AI memory from user information to actual work records. Perplexity's target is also clear. It is demand to improve efficiency in work with high repetition and continuity, such as competitive monitoring, weekly reports and investigative tasks that require continual reference to past execution results. Perplexity emphasised that in such environments Computer "will not reinvent the same wheel every session."

Data control is limited. The context graph, the personal large language model wiki and session records are all stored on Perplexity infrastructure. Users can see what is stored, but the structure does not give them ownership. For that reason, the media outlet also pointed to differences from self-hosted tools that store data on a user's own hardware and allow full control.

Brain can be seen as an attempt by Perplexity to strengthen an agent's ability to carry out continuous work within its own ecosystem. The company said additional features would be coming soon but did not provide a specific timetable.

Introducing Brain in Computer. Brain is a continuously learning memory system. Every task on Computer plugs into a context graph built by Brain. It makes Computer more stateful with every run. Available as a research preview for all Perplexity Max subscribers. pic.twitter.com/Dw4Q7Izmqs