Knitli

NatSec friends: help me find ten (or more) people for a 30-minute conversation. I’m trying to understand my market. No sales. No demo. The conversation is about your experience, not me or Knitli. How does classification compliance software get chosen? What gets built, reused or bought? Who influences and approves the choice, and what requirements constrain it? Think: - Classification-marking forms in web apps, SaaS/PaaS platforms and tools in Outlook/Office. - Classification and compliance auditing. - Network and cross-domain validation. - Records metadata, cataloging, storage and declassification. - Experiences with other compliance-supporting software are useful too. I’d especially like perspectives from: - Government organizations. - Data leaders at cleared contractors or in government. - Integrators delivering to government customers. - Product companies that must record, validate or use classification markings. - Contractors that own or operate SCIFs, including those choosing or building software for their own operations, and companies supplying those tools. You don’t need purchasing authority. Knowing one part of the process is useful, including how user experiences and frustrations reach the people who can change something. I’m also interested in requirements and the consequences of noncompliance: the time, work and costs associated with spills, inadequate validation or marking errors, and who bears them. General descriptions only; please don’t share classified, controlled or other sensitive information. I’d like to hear an example or two from your experience and ask you questions about how things worked. Book 30 minutes: https://lnkd.in/eNHMQ-qa Can’t meet? DM me or email adam@knitli.com. Even a brief perspective helps. Know someone I should hear from? Please repost, forward this, or DM me an introduction or contact. If you suggest someone to approach, I’ll send them a single LinkedIn message. A little context: I left DIA as a senior leader after 18 years to start Knitli. I’m a solo founder working to make the classification system work. I built Marque, a compiler that automates classification markings. Now I’m figuring out how to reach trial users and initial customers, build partnerships, and prioritize the work that would help most.

For anyone in the National Security community, it's not a secret that the classification system is complicated. We think it's worse than that. It might be impossible for people to get right today. https://lnkd.in/eD_MKtbf

Excited to announce our first product! Marque is available now. Marque automates the hard parts of classification markings -- resolving all of the rules across each portion, and banner, and ensuring they all are not only correct, but align to the 'most restrictive' markings. In the NatSec community, this is an accepted cost of doing business that takes mission time, and is easy to get wrong. Marque lets the classifier focus on the classification task and automates the marking task. Built in stealth; not an MVP -- full capability ready for integration. Our new website is live with information about Marque: https://knitli.com Interested in how Marque came to be? Read our founder's account: https://lnkd.in/ey2x3Tnf

Most AI coding tools fail not because of bad models or prompts, but because of poor context management. I call it the "goldfish problem" - your AI forgets what it just learned and has to re-read everything, burning tokens and slowing down with each interaction. At Knitli, I'm tackling this through context engineering. It's the difference between AI tools that actually help versus ones that just rack up API costs. The hard truth: summarization loses critical details. RAG retrieves irrelevant context. Caching helps but isn't a complete solution. Each approach has real tradeoffs, and there's no universal fix yet. What works? Keep sessions focused. Provide explicit context upfront. Match your tools to tasks based on how they handle context windows. Understanding these limitations is the first step to building AI workflows that don't waste your time and money. I built CodeWeaver, Knitli's open-source code search tool, with these principles in mind. It's designed for the Model Context Protocol and optimized to give AI agents the right context without the bloat. Learn more in my beginner friendly deep dive on context engineering and the goldfish problem: https://lnkd.in/eDTYGWR7