--- title: "Every Firm Should Own Its Memory ⬩ Railblocks" description: "Context is all you need. Why a firm’s memory and context should belong to the firm." url: "https://www.railblocks.com/blog/every-firm-should-own-its-memory/" --- Essay 2026.10.27 ⬩ 7 min # Every Firm Should Own Its Memory Eric Friedman Co-founder Figure: Knowledge that stays with the firm. Decisions. Context. History. Your firm owns it. Access it anywhere: chat, email, MCP. We are building Firm OS today with VC and PE firms who already use it. You can see more at [firmos.railblocks.com](https://firmos.railblocks.com). The most valuable thing a firm owns is its memory: its investment decisions, the reasoning behind them, company updates, LP questions, and how things get done. At most firms this institutional knowledge is spread across inboxes, folders and people’s heads. Platforms, people, tools and offices change over time, and the firm should be able to take its memory with it. For years, SaaS apps and knowledge platforms have asked firms to put everything inside them. We think the firm should own it instead. If you use Dropbox, sync it with Firm Memory. If you have years of Gmail history, bring it into Firm Memory. Each tool the firm already uses becomes a source, and the memory stays with the firm. ## Memory was stuck until now Data portability made this possible, and AI makes it usable. Firms could always download their fund admin data, bank statements, portfolio updates, IC memos and founder decks. What they couldn’t do was drop those files into an AI tool, pull out the facts and memories, and then carry that memory over to the next agent or chatbot they wanted to try. The memory existed, but it was stuck. In 2017, a team at Google published [“Attention Is All You Need,”](https://arxiv.org/abs/1706.03762) the paper behind the models firms use today. For a firm, the line that matters now is a little different: **context is all you need.** The models are already good, and they get better every few months. What they don’t have is your firm’s context: why you passed on a deal, what an LP asked last quarter, how the partners think about a sector. Give an agent that context and it does the work the way your firm would. Firm OS lets you build your own Firm Memory, which you run and control, with the building blocks you need and access from anywhere: - **Trying a new agent?** Add your firm’s memory as an MCP. - **Need it in Slack?** Connect an agent in Slack to your firm’s memory. - **Want to use email?** Set up an email address for the firm. Figure: Five source groups feed the Railblocks Hub with Firm Memory on the center block; the Atom represents your agent. Desktop, a mobile phone and your agent through MCP connect to the same firm-owned memory. **You’re no longer beholden to someone else’s platform. Your firm can be its own.** ## Vibe coding got firms started AI coding tools let anyone build and ship what they can imagine in hours or days instead of years, on their own computer. Anyone with Claude Code or Codex can stand up what they need quickly. Meanwhile, large consulting firms, SaaS platforms and dev shops charge hundreds of thousands to millions of dollars for 6 to 12 month AI transformation roadmaps that are hard to act on. Our best clients figured much of this out by vibe coding. They have been building in Claude Code, Codex, Gemini, Amp, Copilot, Ghostty, Herdr, Moshi, OpenClaw, Hermes, Manus and others, dragging in PDFs and connecting Google Drive, inboxes and iMessage. Then they ran into the firm’s policies. IT requirements, RIA rules, the firm’s security posture and data archival requirements quickly outgrow a system running on localhost:3000. A tool one person vibe coded on their laptop can’t be used by the rest of the team, and nobody keeps it up to date. To run a firm on it, you need a platform that is multiplayer, stable, secure and extensible. That’s why we build on Cloudflare. ## Why Cloudflare In 2013, Brad Burnham wrote about [why Union Square Ventures invested in Cloudflare](https://www.usv.com/writing/2013/12/cloudflare/). Rereading it, the reasons are amazingly similar to ours. > “the millions of smaller sites who could never dream of deploying and managing this kind of infrastructure themselves.” Brad Burnham, [“CloudFlare,” Union Square Ventures](https://www.usv.com/writing/2013/12/cloudflare/), December 17, 2013. Excerpt. Big networks bought equipment for performance, load balancing, WAN optimization, firewalls and DDoS defense. Cloudflare turned that edge into a service, and opened the market to the millions of smaller sites that could never run it themselves. Swap “sites” for “firms” and “infrastructure” for “AI,” and the argument still holds 13 years later. What only the largest firms could build, a 10 person firm can now run. In 2013, Cloudflare was early in making the edge available to everyone. In 2026, we are early in the same shift for AI inside firms. ## What we believe 1. Firms need AI leverage. 2. Firms should own their own memory and context. To do this they need to own it and run it on their own infrastructure. The firm is the platform, not another SaaS tool. ## Enter Railblocks Railblocks is a forward deployed engineering team. We embed with your firm and build an AI system you own and control. We connect the systems you already use, put the building blocks you need in place, and set everything up on cloud infrastructure you control. Everyone wants “AI transformation,” but most firms don’t know where to start, how to measure progress, or what good practice looks like for VC funds, PE firms and family offices. We have worked with a handful of firms so far, so we know what works and what doesn’t. We can help you evaluate the tools, services and software contracts you already have, and build solutions this quarter. We start by asking: - “Show us what you want to do.” - “Show us what you built and now want the team to use.” - “Show us how things happen today.” Then we build in short cycles, talk through interactive demos, and ship in days rather than months. If you can vibe code it in a weekend, you should come work with us as an FDE. Most partners have day jobs, though, and someone has to maintain the system every weekend after that. If you want your weekends back, that’s where we come in. ## What we build with you **Memory the firm owns.** Owning your memory doesn’t mean running your own servers. It means the firm holds the keys, decides who sees what, and can export everything and leave at any time. Models will change, and the memory should outlast them. On CNBC in July, Palantir’s Alex Karp said CEOs tell him the labs are “stealing the weights and alpha of my business.” Karp has his own platform to sell, but the instinct is right: whichever model you use this year, the knowledge it works from should belong to you. **Multiplayer Experience.** Everyone on the team works in the same system with the right permissions. What one person teaches it, the whole firm gets. **Security from the start.** We have rolled this out across investment firms before, so you don’t start from scratch. Your system follows the firm’s rules and its regulatory and archival needs, gets pen tested, and works within your IT team’s guidance. It sits behind Cloudflare’s DDoS and malware protection, the same edge Brad wrote about in 2013. **Stability.** It runs every day for the whole firm, not on one partner’s laptop. We set it up, monitor it and keep it current. **Integrations (for everything you use today, and whatever you use tomorrow).** The firm should be the platform. Everything connects through an MCP, an API, the web or a report download, so the firm can switch vendors easily. If your bank doesn’t have an MCP, we set up a worker that pulls the data daily and combines it with a banking partner that does. As SVB showed, having a backup matters. **Embedded team.** Forward deployed engineers who build with your team, keep the system running, and bring what we learn across firms to each client. ## The risks we are solving for **Model risk.** Don’t depend on one frontier lab. **Security risk.** Build on infrastructure that already protects a large share of the internet, not on a startup’s servers or one partner’s laptop. **Firm risk.** Don’t let a long SaaS contract make your decisions. What’s best for the firm may not be what’s best for a sales rep’s quota. ## What every firm gets Railblocks builds a series of blocks, with systems engineers in the background who push updates and build on what each firm needs. **Firm Memory.** The first block, and the first beta of Firm OS. It pulls in what the firm already has (portfolio updates, email, Dropbox, the CRM) and turns it into a vector database of facts, memories and source documents. It runs on infrastructure the firm controls, and the firm can export all of it at any time. Whether you just wrote a thesis or have 10 years of email history, you should be able to point to where that data lives and ask it questions. **Agentic leverage.** Three questions worth asking: 1. Can your people ask questions of your memory system? 2. Can agents help them get work done, or finish work faster and more accurately? 3. Are you building institutional knowledge, or teams where things collapse when someone leaves? ## Own your context and memory Cloudflare gives us the infrastructure. Railblocks is the team that builds on it with you. At the end of the day you have two choices: own it, or rent it. It all starts with Firm Memory. Learn more and start building your firm’s memory at [firmos.railblocks.com](https://firmos.railblocks.com). ### Sources - [Brad Burnham, “CloudFlare,” USV, Dec 17 2013](https://www.usv.com/writing/2013/12/cloudflare/) - [CNBC on Alex Karp, Jul 1 2026](https://www.cnbc.com/2026/07/01/palantir-karp-open-ai-anthropic-tokens.html) - [Business Insider on the same interview](https://www.businessinsider.com/alexander-karp-criticizes-ai-companies-token-costs-2026-7) - [Vaswani et al., “Attention Is All You Need,” Jun 12 2017](https://arxiv.org/abs/1706.03762) [ Newsletter ] ## The secrets behind the best-run teams Discover the tools, systems and strategies top companies use to streamline their entire operations with AI.