Verascient Raises $1.2M for Time-Aware, Permission-Aware Enterprise AI

Cape Town-based Verascient has raised an oversubscribed $1.2 million first round to build out enterprise AI deployments. Its approach centers on a temporal knowledge graph: a record of organizational information that the company says retains history, permissions and provenance for AI agents.

Founder Collective led the round, joined by Andrena Ventures, Cambridge Enterprise, Summit Ventures and angel investors. Alan Knott-Craig and Shayne Mann participated as angels. Verascient says it will use the capital to expand its engineering team, deepen its technology and support more enterprise deployments.

Both accounts put the raise at $1.2 million, but convert it to slightly different rand amounts: R19.5 million in one and about R19.2 million in the other. The dollar figure is the consistent measure of the round.

Turning scattered records into agent context

The product takes in company knowledge spread across documents, emails, meetings, spreadsheets and internal systems. Verascient consolidates that material into infrastructure for AI agents, while its temporal graph is intended to preserve the record’s history, access permissions and provenance, or where information came from.

That design is meant to make the knowledge layer an organizational record rather than a simple collection of files. Ferreira says it maintains a full picture of an organization while making information available to agents; the time-aware element is the retained history, alongside permissions and source lineage.

The mechanism addresses a deployment problem Verascient identifies: businesses may give employees AI tools without building systems that allow the company itself to operate intelligently with AI. The company says it works inside client organizations to understand people and processes, then places agents within existing systems and access controls.

Verascient says it pairs that infrastructure with in-house AI engineers who work with client teams to identify uses and build systems around business requirements. The company also says its platform has more than 1,000 integrations and is designed for existing business systems and access controls.

A focus on information-heavy businesses

Verascient is initially targeting financial services, insurance and logistics. Those sectors commonly hold large volumes of institutional knowledge across disconnected sources, a setting where the company aims to organize information and use it in AI-supported workflows and agents.

Co-founder and CEO Keagan Stokoe has said the company starts with work tied to revenue, operations, customer experience, decision-making and delivery. That focus places the technology in day-to-day business processes rather than treating it as a standalone AI tool.

Stokoe’s stated thesis is that each operational improvement makes the next easier to implement. That is a company view, not a disclosed customer outcome, but it explains why Verascient is combining integration work, workflow design and agent deployment rather than selling a disconnected AI interface.

More builders, with the proof still ahead

The founders bring relevant operating experience. Emile Ferreira was an early developer at Replit, built an on-device AI assistant and completed an MPhil in Advanced Computer Science at Cambridge. Stokoe was part of Fibertime’s founding team before establishing AI consultancy Purple Dorm.

Ferreira’s on-device assistant drew more than 200,000 users and won an innovation award while he was in high school, according to the company’s account. Stokoe’s prior consultancy worked with organizations in South Africa and the UK, while the founders say Verascient intends to use South African talent to serve international markets.

The immediate hiring plan is for a small team of engineers and AI builders; Stokoe has described the recruiting goal as the top 1% of AI talent in South Africa. The announcement does not provide customer counts, deployment totals or independently measured performance, so the commercial results of that model remain unquantified.