1:00 PM Eastern (EST) / Hosted by David Braun
The second Show & Tell had two demos. David showed workflowtech.ai, where the front door is a terminal that answers from CODEX and cites its sources. One live question came back "cannot find", which turned into a look at how the retrieval works and what it still misses. Raum Sandoval showed Flowchestra, a secure AI operations platform that helps MSPs turn their clients' demand for AI into managed AI services, starting from an intake that maps the business before any AI goes in. Afterward the talk turned to configuration drift, and to why MSPs still lack one view across tools that each only see their own domain.
Demos
workflowtech.ai: a terminal that answers from CODEX
David Braun
David's company site has a terminal as its front door instead of a contact form. Four intake questions pick one of three personas, and a router re-picks on every message. Answers that come from CODEX, EGI's automation library, are held back until a check confirms they cite a source. CODEX started as an archive of about 50 Rewst community repos and now refreshes every night from sources David picked, including Anthropic, OpenAI, Simon Willison and r/ClaudeAI. A live question about the 2026 State of MSP Automation report came back "cannot find"; the terminal has no web search by design, so it says when it can't find something instead of guessing. David also covered two lessons from building it. A keyword search passed every test and did nothing in production, because the test library was tiny and production held about 63,000 chunks. Moving the keyword index onto disk with SQLite FTS5 fixed it, and measured on the real corpus, the right passage now lands in the top three about half the time, which the whitepaper calls a floor. And a router sent "integrate" to the sales persona because the word contains "rate".
Demo, 4:16
Flowchestra: secure AI operations for MSPs
Raum Sandoval
Raum has spent about 20 years in the MSP space, most recently as Chief Innovation and Information Security Officer at an MSP whose clients were mostly regulated financial-services firms. Flowchestra is a secure AI operations platform built for MSPs. It starts with an onboarding intake that maps a client's strategy, compliance needs, risk tolerance, teams, processes, tools and metrics, and uses that to shape both the experience (workspaces and recommended use cases) and the controls. Tools connect through APIs and MCP. The guardrails set which models and providers are allowed, can require zero data retention, and flag, log, tokenize or substitute sensitive information, with policies set per workspace or per person. It is multi-tenant. Next is a blueprint that maps who is responsible for risk, value and approvals across the organization.
Takeaways
- Map the business before you deploy AI: roles, risks, goals and tools come first.
- Securing AI is where most MSPs are focused now. The longer-term opportunity is in understanding the client's business.
- Measure retrieval on the production corpus. A change can pass every test and do nothing in production.
- You can know every backup ran and still not know whether every machine that should be backed up is in the system.
- Hold answers back until they cite a source, and when the library doesn't have it, say so.
Joke of the month
What was Obi-Wan Kenobi's advice to Jedis with computer problems?
Punchline
May the force-quit be with you.
Tool of the month
A universal semantic layer: one SQL query runs across distributed data sources as a single data fabric. David used it for a recent hackathon build.
Recording
Each demo gets its own video above. The full call isn't posted.
