# Coast records desktop context locally to give AI agents persistent memory > Source: > Published: 2026-07-28 19:00:59+00:00 [Attention Engineering](https://attention.inc/) co-founder [Aidan Guo (@aidangch)](https://x.com/aidangch) opened a local preview of [Coast](https://coast.app/) on July 28th, pitching software that records desktop context on-device so users and AI agents can retrieve it later. [https://x.com/aidangch/status/2082134857862578585?s=46](https://x.com/aidangch/status/2082134857862578585?s=46) In a [29-post thread on X](https://x.com/aidangch/status/2082134857862578585?s=46), Guo framed daily computer activity as a valuable dataset that disappears when people finish working. Coast is Attention Engineering's attempt to preserve that context, giving agents a persistent record instead of forcing users to reconstruct their work in every new session. Attention Engineering calls Coast "fully local memory for you and your agents." Guo described the preview as "on-device perfect memory," although Attention Engineering has not published evidence supporting the absolute performance implied by that phrase. The immediate product is a memory layer. Attention Engineering's broader plan, described in 2025, is an ambient desktop assistant that learns users' workflows and eventually automates repetitive tasks across applications. ([coast.app](https://coast.app/)) Guo's route to the product started well before AI agents became the center of enterprise software. On his [personal site](https://guo.so/), he says he grew up in Vancouver, began botting sneakers at 13 and left Carnegie Mellon computer science after high school to move to the Bay Area. He previously ran footwear resale group PremeProfits and co-founded sustainable stationery business Fern. ([guo.so](https://guo.so/)) Guo founded Attention Engineering in 2025 with [Julian Windeck](https://julianwindeck.de/), a German computer scientist who studied at Cambridge and conducted AI research at MIT and Berkeley. The pair met through Silas Alberti of Cognition and Marvin von Hagen of Interaction, worked together on smaller projects and then formed Attention Engineering. Windeck's research and systems background gives the founders a technical complement to Guo's product and fundraising role. ([julianwindeck.de](https://julianwindeck.de/?utm_source=openai)) ### Memory as the wedge Coast narrows Attention Engineering's original desktop automation pitch to the prerequisite every agent needs: context. An agent can generate text or operate software, but useful action depends on knowing what the user has already seen, decided and done. Attention Engineering is betting that the business controlling that record can sit underneath multiple agents rather than competing for every user interaction. Attention Engineering's hiring materials show the infrastructure behind that bet. An open [member of technical staff role](https://jobs.ashbyhq.com/attention/d6749927-d4e9-4e37-819a-0efeb28e8180) covers agent orchestration, context management, visual language models, search, retrieval, vector systems, evaluation and workflow architecture. That is the stack required to turn raw activity into information an agent can retrieve at the right moment. ([jobs.ashbyhq.com](https://jobs.ashbyhq.com/attention/d6749927-d4e9-4e37-819a-0efeb28e8180?utm_source=openai)) The preview also marks a change in what Attention Engineering can ask users to adopt. A general desktop agent must prove it can act reliably across many applications. A memory product can deliver value earlier by helping users recover past work and supplying context to other software. Each day of captured activity can make Coast more useful while increasing the cost of switching to another memory system. ### The trust test Coast's central product decision is also its largest risk. Software that observes enough desktop activity to understand a person's work may encounter private messages, customer records, source code, financial data and credentials. Attention Engineering is putting its on-device architecture at the center of the pitch because the product requires a level of access that users are unlikely to grant casually. The trade is direct: Coast must capture enough context to become useful, while users must trust Attention Engineering's claim that the resulting memory remains local. Controls over recording, retention, deletion and agent access will determine whether persistent memory feels like infrastructure or surveillance. ### Backed to build the context layer Attention Engineering raised $1.25 million in pre-seed funding in 2025 from backers including Lukas Haas, Village Global, Liquid 2 Ventures, Bryan Pellegrino, Marvin von Hagen and Felix Schlegal. Reporting at the time said Attention Engineering planned to spend the capital on machine learning, systems, privacy work and early enterprise pilots. ([businesscloud.co.uk](https://businesscloud.co.uk/news/cambridge-graduate-launches-ai-firm-with-teen-in-san-francisco/)) That investor group funded a broader vision than the preview Guo released on July 28th. Attention Engineering previously described an assistant that would anticipate actions and automate desktop sequences. Coast starts one layer lower, building the personal data store such an assistant would need before it could act with useful context. The launch puts the burden on retrieval quality and user control. If Coast can reliably recover the right detail without sending sensitive work off-device, Attention Engineering can become a shared memory layer for desktop agents. If the archive becomes noisy, resource-intensive or difficult to govern, the same recording that powers the product will limit its adoption.