praxic watches how your team works, finds the automation opportunities they can't see themselves, and — with the employee's approval — builds them into skills the company owns. So institutional knowledge stops living only in people's heads.
Currently accepting design partners for teams of 20–200.
Every company runs on workflows that exist only in the habits of its people. The operations manager who exports the same report every Friday. The customer success rep who copies information between three tools after every call. The engineer who writes the same status update every sprint. These patterns are real, they recur constantly, and they are completely invisible to every documentation system, wiki, or AI tool you have tried.
The only way to find them is to watch how people actually work. Not what they say they do. Not what the wiki claims. What they do.
Every AI tool on the market waits for you to tell it what to do. You write the prompt. You define the task. You do the cognitive work of recognising what needs automating.
praxic inverts this entirely. Every action your team takes is a prompt. Every app opened, every sequence repeated, every Friday-afternoon ritual — these are behavioural signals. praxic reads them continuously, finds the patterns, and surfaces the automation opportunity to the employee at a natural break in their day.
The prompt is the work itself. You just have to be watching.
Three layers, one continuous loop — from opt-in observation to an approved skill that runs without anyone asking again.
A lightweight agent installs on the employee's machine with explicit opt-in permissions. It captures behavioural metadata — app sequences, action patterns, timing — and the named targets of an action, like which report or which recipients. Never the content of messages, documents, or files.
Once a workflow has recurred enough times to prove itself, praxic surfaces it at a natural break in the employee's day — in plain language, with the evidence behind it: how many times, over how long, at what time of day. They approve, edit, or dismiss.
Approved workflows compile to a skill. The employee watches a supervised first run before it may run on its own; every execution after that is audit-logged. Skills join the company's shared library, built to the Anthropic Agent Skills open standard — portable, auditable, company-owned.
Every “company brain” product you have seen indexes your documents, drops them into a retrieval pipeline, and calls it institutional knowledge. That fails for a structural reason: the most valuable knowledge was never written down, and the documents that were are contradictory, outdated, and unmoderated.
praxic derives knowledge from behaviour, not text. From what people actually do, not what the wiki says they should do. That is a categorical difference in the quality of knowledge that comes out the other side.
“The most valuable operational knowledge in any company has never been in a document. It lives in the sequence of actions a person takes without thinking about it.”
Every skill praxic creates is compiled to the Anthropic Agent Skills open standard — now supported by Atlassian, Notion, Figma, Zapier, and others. Integrations are handled through the Model Context Protocol (MCP), now under the Linux Foundation. Your skill library works across any compliant agent platform, not just ours.
The agent observes how work happens — apps, sequences, and the names of things you act on — so your skills set themselves up. It never reads message or document content, never logs keystrokes, never captures screenshots, never touches credentials. An on-device filter blocks sensitive apps before anything leaves the machine.
Raw observation data is kept only long enough to detect your patterns and build your skills — 30 days, then it's deleted. The skills persist; the raw record doesn't. It is never sold, shared, or reported to any third party.
Observation is opt-in. A real-time dashboard shows exactly what is stored about you, per app, per day, with the deletion clock on every entry. Deletion is one click. Managers see the skill library — never individual data.
Co-founder & CEO
Waterloo. AI Strategy Consulting at Microsoft, where she analysed Copilot's competitive position across the North American SMB market. Software Engineering intern at Muskoka Woods. Building AI products.
Co-founder & CTO
Waterloo Software Engineering. Y Hacks winner. Software Engineer at Ciena. Building the observation agent and pattern recognition pipeline.
We're looking for design partners — teams of 20–200 who want to build this from the beginning.