Archestra Inc., a U.K.-based startup whose open-source platform brokers access between artificial intelligence agents and sensitive enterprise data, today announced that it has raised $10 million in new funding to expand deployments inside large enterprises and grow its open-source ecosystem.

The platform is in production at four Fortune 500 companies, where agents are working across legal, procurement, management and operations. Archestra says enterprise AI teams can wire agents to internal data in days rather than months and without the data exfiltration and prompt-injection risks that have stalled earlier pilots.

Archestra acts as a middle layer between the agent and the data. The agent calls Archestra, which checks identity and policy, brokers the actual data access and logs the result. The company describes it as a horizontal layer that lets enterprises run any popular agent framework or model without rewiring data access for each one. Connectors ship for Jira, Confluence, GitHub, Notion, SharePoint, Google Drive and Salesforce, and the platform supports Claude, ChatGPT, Gemini and open-source models.

The platform runs on Kubernetes and ships with a Helm chart for production deployments and a Terraform provider for infrastructure as code. Secrets sit in a HashiCorp Vault or Kubernetes Secrets with automatic rotation. Model Context Protocol servers run inside isolated containers with per-team access control and full audit trails. Archestra says 95th-percentile latency comes in at 45 milliseconds.

Metrics export to Prometheus, tracing runs through OpenTelemetry and Grafana dashboards ship prebuilt for token usage, request latency and tool calls the guardrails reject.

Retrieval-augmented generation runs inside the platform rather than against a separate vector database. Archestra handles chunking, embedding, hybrid search and reranking in-process.

The company also operates a private MCP registry that lets administrators curate, version and roll back which Model Context Protocol servers are available to agents, drawing on a public catalog of more than 900 evaluated servers. A cost-optimization layer tracks spending per team, agent or organization and can route simpler queries to cheaper models, which Archestra claims can cut inference bills by as much as 96%.

The guardrails are pitched as protection against the combination of access to private data, exposure to untrusted content and the ability to communicate externally that has driven prompt-injection data leaks in ChatGPT, Google Bard, GitHub Copilot, Microsoft Copilot and Slack AI over the past three years. The open-source core has drawn more than 3,700 stars and 57 contributors on GitHub.

“Until now, AI teams had to choose between harshly restricting AI agents and building truly agentic systems, thereby exposing the business to enormous risks,” Chief Executive Matvey Kukuy said. “Archestra.AI eliminates that compromise. Our platform guarantees 100% protection against data leaks while actually accelerating agent development by simplifying data access, deterministic guardrails, governance and observability.”

The new funding will fund engineering and go-to-market hiring, expansion of the open-source ecosystem and faster deployments into regulated industries.

The seed round was led by 20VC, the firm founded by podcaster and investor Harry Stebbings. Other participants include 20 Product, Visible Ventures, Tenacity Capital, Commit Fund, Datadog Inc. CEO Olivier Pomel, HubSpot Inc. Senior Vice President of Agentic GTM and Systems Kieran Flanagan and Permira board member Carolyn Everson.

Today’s round brings total funding to $13.5 million. Archestra raised a $3.3 million pre-seed round in August 2025, when the company was positioning itself as a security layer for the then-nascent MCP.

Photo: Archestra