Supervision and specialization layer that turns production agent traces into evaluated, owned models

Overmind Technology was founded in London in 2025 by chief executive Tyler Edwards, chief technology officer Akhat Rakishev, and chief revenue officer Sam Brunt. Edwards spent eight years building AI systems inside British intelligence, across MI5, MI6, and GCHQ. The company raised a seed round of roughly two million pounds in February 2026 led by Osney Capital, with 14Peaks, Portfolio Ventures, Endurance Ventures, and returning investor Antler participating. It targets legal, healthcare, financial services, and government buyers, where agentic systems carry real regulatory and data protection exposure and sending every request to a third-party frontier model is a governance problem rather than a convenience.

The platform starts by building an agent context graph, discovering prompts, tool definitions, decision paths, and dependencies from the codebase and validating that picture against real production traces, which gives teams an accurate account of what their agents actually do in the field. Those traces are then curated into training and evaluation datasets, with each example checked against its source agent and traceable back to the run it came from. Overmind derives evaluation criteria from the agent's own design, and proposes rewrites to prompts and tool definitions as pull requests scored against baseline performance so changes are reviewable rather than opaque. Teams can also train smaller specialized models on their own data and either serve them from Overmind endpoints or download them for self-hosted inference.