# Hyperspell Review: A Company Brain Built for Other Agents > Source: > Published: 2026-09-09 14:00:00+00:00 # Hyperspell Review: A Company Brain Built for Other Agents Hyperspell calls itself your company brain, which is the same phrase we use, so this is a review written by someone with an interest. Read it that way. Details below come from the company’s own site, read on September 9, 2026. ## What Hyperspell is A context layer. It connects your tools, builds what it calls a context graph, and then, in its own words, “surfaces it as a filesystem any agent can read.” That last clause is the whole product, and it’s easy to skim past. Hyperspell is not an assistant you talk to. It’s the knowledge layer underneath somebody else’s assistant. Their own how-it-works section ends with “plug into Claude Code, Codex, Cursor, your custom agents, internal tools, or query it directly,” and the feature list promises compatibility “with every agent framework, or build your own with the universal API and SDK.” So the buyer is a team that is already building agents and has discovered the model doesn’t know anything about their company. ## The problem they describe Their framing of it is good, and it is close to ours: Your company runs on knowledge buried in Slack threads, Linear tickets, Google Docs, meeting transcripts, and leaves when the employee does. And on why retrieval alone isn’t enough: “Today’s best AI tools retrieve text fragments. They don’t understand how information connects, what changed yesterday, or what actually matters to the task at hand.” That distinction is worth holding onto whoever you buy from. Search over your documents answers “where was this written down.” A model of your company answers “what is true now,” which is a different question, and it’s the harder one. ## How it works Three steps on their site. **Connect.** Slack, Gmail, HubSpot, Notion, Linear and the rest, with more than 50 pre-built connectors. Authentication is OAuth, and their line is that permissions come across with it. Whoever asks a question gets back only what they could already open in Slack or Notion themselves, so nobody has to re-create each tool’s access rules inside the app being built. **Synthesize.** Continuous rather than scheduled. The claim is one bespoke model of the company that stays accurate in real time, and that relevance improves with each query. **Serve.** Structured results or markdown summaries ready for a model, consumed through the API, an SDK, or a coding agent. The permission-inheritance point deserves attention if you are comparing this category. A shared company brain only ships if it can answer differently for different people, and several products here are still on one toggle between private and company-wide. ## What it costs Unknown. There is no pricing page, every call to action is a demo booking, and nothing on the site names a number or a free tier. That’s a sales-led motion, which usually means the price depends on connectors, volume and seats, and that you’ll find out on the call. ## Security SOC 2 certified and GDPR compliant, both stated on the homepage. They store “summaries of that data and extracted memories in our memory network.” A summary is still your content, on their servers. ## Who it is for **Hyperspell fits** if you are building agents and the missing piece is company context. You have engineers, you’ve already wired up a model, and what you lack is a governed, permission-aware view of Slack and Notion and your CRM that an agent can query. Buying that layer rather than building it is a reasonable call, and one of their own testimonials makes that argument: assembling ingestion, search and monitoring “would have taken months.” **Hyperspell does not fit** if you want software your team can use right away. There’s no inbox, no task list, no place for a non-engineer to go. The output is an API response, and turning that into something a salesperson or a founder touches is your project, not theirs. ## Where we fit We build [this+that](/), and we sell to the second group. Our [Brain](/brain/) holds the same material, operational knowledge assembled out of the work rather than written by hand, and workflows write back to it as they run. The difference is what sits on top of it. We ship the surfaces too: an inbox across every channel your team uses, [DoBox](/dobox/) turning what arrives into assigned tasks, and [agents](/agents/) and workflows acting on it. Nobody has to build the product around the knowledge layer, because everything is built into our product. Hyperspell’s continuous synthesis is ahead of where we are on zero-configuration currency, which for us is still workflow-driven write-back rather than automatic upkeep of everything. We run an MCP server too, though it points the other way. Ours lets Claude or ChatGPT work inside this+that, reading and writing tasks and brain pages, rather than serving as a context layer for someone to build a separate product on. The overlap is the premise, that a company’s knowledge should assemble itself. The split is who does the assembling of the product around it. ## Frequently asked questions ### Is Hyperspell an assistant? No. It’s a context layer that other agents read, through an API, an SDK, or a coding tool like Claude Code or Cursor. ### How much does Hyperspell cost? The site publishes no prices and every call to action books a demo. ### How current is the data? They describe continuous synthesis and say the model stays accurate in real time, rather than refreshing on a schedule. ### Does Hyperspell handle permissions? They say a source’s own access rules carry over when it’s connected, so a person retrieves only what that tool would already show them. Their wording is “OAuth in, permissions inherited automatically.” ### Is Hyperspell secure? The homepage claims SOC 2 certification and GDPR compliance. They store summaries and extracted memories rather than just pointers.