We design, deploy and scale production AI for enterprise teams — private workspaces for the whole company, custom agents inside your operations, and AI features embedded in your product. End to end, owned by you.
Open-source labContextFence — test RAG permissionsWorked with
Workspace · Finance · 4 messages · cited
Summarise the Q3 board pack. Three bullets. Plain English. Cite the doc.
Revenue up 12.4% YoY, driven by the new EMEA region launching ahead of plan.
Operating margin compressed 180bps — supplier costs and the new product line ramping more slowly than forecast.
Cash position healthy; refinancing window opens Q1.
Self-hosted·SSO enforced·Audit logged·Data stays in your tenant
What we do
Most clients arrive with one of these in mind and discover the others over time. Pick a starting point — we'll grow from there.
A secure, multi-model AI platform for your whole company — self-hosted, governed, and rolled out properly.
Agents and copilots woven into the systems your team already uses — drafting, triaging, summarising real work.
Embedded AI features in the platforms you ship to customers — search, vision, generation, at scale.
Workshops, audits and prototype sprints for leadership figuring out where AI moves the numbers.
Devector labs
Alongside client work, our labs explore two ends of product engineering: reliable AI systems and playful mobile experiences built for the real world.
Playwright for RAG permissions.
ContextFence tests declared identity × source boundaries as deterministic regression checks. It helps catch a shared cache, stale ACL, or bad retrieval filter before it reaches a real user.
Maya · Newsroom
Asks for a margin forecast
Finance Vault
Excluded from retrieved context — as expected
Deterministic assertion
source_absent: finance-vault
A solo, real-world deduction game played on a city's transit network.
Pick where you are, reveal one useful clue, then follow the live map to catch a hidden station. No account, no lobby, and no raw location trail sent to a game server.
Let's talk
Most engagements start with a 30-minute call. We'll talk about what your business needs from AI, walk you through a working deployment, and tell you honestly whether we're the right shop for the job.
