past.dev

past.dev ranks first on BEAM at every history size from 100K to 10M tokens, and passes 490 of 600 DolphinBench tasks. Full results, the evaluation method and every source.

The past.dev marks, colors, type, and usage rules, with downloadable SVGs.

The page describes how context engineering determines the instructions, tools, evidence, history, and state included in each model call, including guidance on managing and measuring context.

SOC 2 Type II, ISO 27001, ISO 27701, GDPR: the certifications, the engineering controls behind them, and how to get the reports.

Engineering, research and company notes from past.dev, with every post readable as Markdown.

What shipped each week in the past.dev Memory API, the console, the MCP servers and the integrations.

Problem-space guides for teams building AI agents: changing facts, audit trails, retention, erasure, and source-backed recall, each answered directly.

past.dev, a memory layer for AI agents, is available today. An agent asks it for context and gets current facts with dated sources, limited to what it may see.

past.dev can be deployed as a managed service or in a user’s own cloud account, using the same engine and API in both cases.

The company offers three plans priced by credits, with one credit covering 350 bytes of ingested payload. Recall costs apply only to the Pay as you go plan, while storage, entities, and seats are not metered.

past.dev is a mem0 alternative for temporal storage. Compare evidence, deployment, changed-fact handling, and tests for evaluating either system.

Compare Letta and past.dev by agent runtime, memory control, ingestion model, deployment, and recall output. The products can also be used together.

Compare past.dev and mem0 by storage model, input format, deployment, changed-fact handling, recall output, and operational requirements.

Supermemory provides connectors and multimodal ingestion. past.dev accepts timestamped text and returns ranked documents with dated source excerpts.

past.dev is a small team in San Francisco building long-term memory for AI agents, founded by Mehdi Djabri and Alex Ronse.

Email contact@past.dev to reach the past.dev team, support@past.dev for a product question, or book a call about integrations, Enterprise and security. Signing up needs no call.

Long-term, multiplayer memory for agents. The long-term memory API, #1 on BEAM, the largest public memory benchmark.

The cookies and browser storage past.dev uses, who sets each one, why, and for how long.

What the past.dev website and Memory API store, who can access the data, whether it trains models, and how to request deletion.

The MCP servers use the same security and privacy controls as the rest of past.dev.

The page outlines the terms for past.dev, including the Memory API, console, and MCP servers, covering accounts, acceptable use, content policies, model output, fees, credits, termination, and liability.

GitHubRepository owner

Past.dev’s DolphinBench achieves the highest score for AI agent memory, improving how agents retain and update past interactions. The benchmark evaluates agents’ ability to incorporate new information while maintaining accuracy in long-term recall. Developers cite its role in advancing AI systems for dynamic, context-aware applications.

Oct 2026AI-Tech Park - AI, ML, IoT, Cybersecurity News & Trend Analysis, InterviewsNews article

past.dev achieves the highest accuracy score on DolphinBench, a benchmark for AI agent memory retention. The platform improves how AI agents recall and process information across long conversations. This advancement addresses challenges in maintaining contextual accuracy as interactions grow.

Oct 2026PR NewswireNews article

past.dev passes **490 of 600 tasks** in DolphinBench, scoring **81.7%**—the highest benchmark result reported to date. The performance uses **Hermes Agent** alongside **GPT 5.6 Luna**. The benchmark evaluates task execution across diverse scenarios, positioning past.dev as a leader in agentic AI workflows.

Oct 2026past.devNews article