In a typical product team, context and key decisions live everywhere across Granola call notes, Linear tickets, Figma files, Slack conversations, Notion docs and Github. Teams lose time changing tabs to answer basic questions about why something exists, and that context switching can be tough mentally, impacting the speed of delivering a project.
Typically wikis, knowledge bases and onboarding documents are created to maintain shared project context, but they rarely help. Most end up rotting in the background with teams lacking the time to keep them up-to-date manually, meaning they lose value until they’re forgotten entirely.
Enter Warp, a tool we built to maintain this shared context. An LLM maintains a queryable living ‘brain’ for every Loomery project, tracking everything about an engagement, bundling tools via MCPs and performing the tedious task of ingesting and updating the brain async, leaving the humans on the team to focus on the more complicated delivery work.
What is Warp? Warp is a version-controlled repo that pulls together context that would otherwise live scattered across sources like call notes, tickets, designs, decisions, docs and code.
Instead of hunting through Slack threads or asking around for specific context Warp lets us:
Own the context: The repo lives on GitHub with full version history. If anyone on the team clones it once, an AI coding tool inside that folder has shared project context automatically (who's who, what's decided, what tools are relevant). The dashboard provides a top-level view of the latest developments, where the project sits on the timeline and who's working on what. Reach it from anywhere: Register the repo as a plugin marketplace so /<project-name>:brain works from Claude Chat, Cowork or a different project, not just inside the repo's own folder.Keep it current without extra effort: An ingestion ritual means every new Claude session checks all sources for anything new and proposes updates, reviewed and approved by a person, not auto-committed.We've built a mock version you can explore , based on a fictional project, to see the brain in action.
The result? An engineer can ask about a feedback session, a release or the design system without leaving their editor. A PM prepping for a stakeholder update can instantly find out what's changed since the last one, instead of trawling Slack and Granola. A new joiner to the team can access an up-to-date history of everything that has happened on the project and can be productive from day one.
How does it work? A new brain is fired up using a script that generates the initial scaffolding for the brain. This spins up a new GitHub repository bundling two generic skills out of the box: /brain and /onboard. The latter skill is used once to get the agent to help seed the brain with relevant sources, context and client tools.
The resulting repo is structured in three layers:
Agent instructions: these include (1) AGENTS.md which is always loaded covering quick facts, repo layout and hard rules, (2) CLAUDE.md, a one-line pointer for Claude Code, and (3) a single shared playbook every harness’s adapter points at rather than duplicating instructions per tool.Curated project knowledge: these are human-readable markdown summaries covering everything we know about the project including people, goals, decisions, meeting logs and key concepts.Live sources: the fast-moving and large material, like transcripts, chat history and drives, aren’t copied in. The repo holds an index of pointers and Claude fetches the real thing on demand via MCP connectors to things like Granola, Slack, Linear, Miro and GDrive. A per-project whitelist file scopes exactly which folders, files and tools an agent can pull from. This matters because the underlying Drive and tools are often shared infrastructure across clients, so the wrong data never gets pulled in. Client code is the exception, it’s brought in directly as a git submodule, not fetched live.Every new Claude session checks all sources for anything new and proposes updates meaning the brain’s knowledge compounds over time, staying current as a side effect of normal working, without requiring a separate maintenance task.
Beyond answering questions about a project, Warp’s consistent access to up-to-date context enables it to help our team write code, make architectural decisions, plan delivery and even write back to tools (always with a human in the loop!). It supports real-time trade-offs our teams are making, like simplifying a feature to launch sooner, because the context to make that call is close to hand.
Helping us move fast Consider a real instance where one of our engineers asked their project brain to explain the context behind a past incident where one user flooded the system with thousands of files at once. An engineer needed to know what happened and to assess how recent code changes would handle the same situation differently.
That question needs incident history, the original decisions and the current implementation all at once. The information existed but was spread across the original discussion in a Granola call, resulting in work in Linear, designs in Figma and implementation in GitHub. You won't get that from reading code alone, and answering it the old way meant finding the relevant piece of information in each place and connecting them by hand. Warp pulls all of those threads together at once, closing the information gap which previously would have required a ‘librarian’ to manually gather.
That’s just one example, our teams are increasingly using Warp to move faster. And once the partnership ends, we transfer the brain repository to the client, so everything the team learned stays with them.
One tool in our move to become AI native We don’t think AI should just mean doing more, faster, be that writing more code, drafting user stories faster or turning around design iterations quicker. Across the whole team, its greatest strength is in giving us more time to think by offloading time-consuming tasks so bandwidth goes towards higher-level thinking. Warp is one part of how we're building that muscle.
The name, of course, comes from weaving. A warp is the set of threads that give a loom its structure, the backbone the rest of the fabric gets woven into. Fitting, for the thing holding a project's structure together while everything else gets built on top of it.
If you're interested in how Warp could accelerate delivery for you, get in touch .
