Transforming AI Agents with Frontier Memory
On October 6, 2026, past.dev unveiled its latest technological advancement—a sophisticated memory API designed specifically for AI agents. This innovation empowers AI to maintain accurate long-term memories, enhancing their performance when navigating complex environments such as corporate communications, meeting transcripts, and digital documents. In a realm where data accuracy and relevance are paramount, past.dev is setting new standards by achieving top rankings on BEAM, the largest public memory benchmark.
What is Frontier Memory?
Frontier Memory enables AI agents to track a variety of company data, including emails, chat messages, meeting notes, and more. The API ensures that each piece of memory is timestamped, allowing agents to determine not just which information is current, but also what previous fact was replaced and who is authorized to view this information. This level of contextual awareness is crucial for companies that rely on AI for decision-making processes, where the most current and accurate information is often needed.
Unprecedented Benchmarking Success
In an impressive demonstration of capability, past.dev's API scored an average of 85.03% across various memory sizes, surging past the previous best of 68.0% on 10 million tokens. The company excelled in other metrics as well, with respective scores of 92.08% at 100K tokens, 89.63% at 500K, and 90.65% at 1 million tokens. These results underline past.dev's strong commitment to advancing AI memory technology beyond traditional search functionalities.
How It Works
The past.dev API alters the way AI agents manage information. Unlike conventional methods of using entire conversation histories as prompts or relying on markdown documents, this innovative API intelligently differentiates which facts are still true. This capability is vital, as it addresses the concern many users have about data accuracy over time. It utilizes a combination of timestamped text storage and identity-aware querying, which together provide tailored responses based on each user's profile and permissions.
Statement from Leadership
In discussing the significance of this technology, Mehdi Djabri, CEO and co-founder, emphasized the shortcomings of traditional memory models. "For two years, the industry has tried to solve memory with search. But search finds what looks similar, not what's still true," he remarked. Furthermore, CTO and co-founder Alex Ronse pointed out the challenges of determining which facts have supplanted others. He stated, "Deciding that one fact replaced another, across sources, without leaking a private email, is the hard part. We've opened our evaluation harness so anyone can check our work."
Enhanced Privacy and Security
In an era where data privacy is more critical than ever, past.dev adheres to rigorous standards. The company is SOC 2 Type II audited and holds ISO certifications 27001 and 27701. Importantly, none of its models are trained on customer data, ensuring that sensitive information remains secure under all circumstances.
Availability
The new memory API is now available for developers, with an easy self-service sign-up process. It’s compatible with any AI model and accommodates various deployment scenarios, including self-hosting and dedicated enterprise solutions.
About past.dev
Founded by a team of innovators with a mission to enhance AI capabilities, past.dev is a San Francisco-based company specializing in frontier long-term memory solutions for AI agents. Backed by prominent investors including General Catalyst, Varsity, and Connect Ventures, past.dev continues to redefine the landscape of AI technology, ensuring that agents can accurately recall and utilize information that is relevant and secure.
As artificial intelligence becomes an increasingly integral part of various industries, innovations like past.dev’s Frontier Memory API promise to transform the capabilities of AI agents—and in turn, the way businesses operate. For more detailed insights and information about their API, visit
past.dev.