Archestra.AI
A London startup building an open-source broker that checks identity and policy before AI agents can touch internal systems or confidential data.
Visit Website ↗ + Add to Compare Claim This CompanyOverview
Archestra.AI builds an open-source platform that sits between AI agents and internal company systems. When an agent requests data or an action, Archestra checks the request against identity and policy rules, brokers the access, and logs what happened — aiming to let employees deploy agents against legal, HR or other confidential systems without the data-leak risk of ungoverned access.
Founded in London around 2024, the company raised a $10 million seed round in June 2026 led by 20VC, with participation from 20 Product, Visible Ventures, Tenacity Capital and Datadog CEO Olivier Pomel as an individual investor, bringing its total funding to about $13.5 million after an earlier, smaller pre-seed round. The company says its platform is in production at multiple Fortune 500 companies, a claim this research could not independently verify.
Its differentiator is the open-source distribution model for agent-to-system access brokering, a different go-to-market approach than the typical closed enterprise security product, in an AI-agent-security space that is now drawing many new entrants.
Innovation Matrix Assessment
Shipped an open-source identity/policy broker and grew from a small pre-seed to a $10M seed round within roughly a year.
Addresses a concrete and growing CISO concern: letting employees use AI agents against sensitive internal systems without ungoverned data exposure.
Backing from 20VC and a personal investment from Datadog's CEO are real signals, and the company claims production use at multiple Fortune 500 companies, though that specific claim is self-reported.
The open-source approach to agent-to-system brokering is a genuinely different distribution model from typical closed enterprise security tooling, though the broader AI-agent-security space is now crowded with well-funded competitors.
The company's own claim of guaranteeing complete protection against data leaks is explicitly a marketing assertion; no independent testing or named-incident evidence was found.
As enterprises deploy more autonomous agents against sensitive systems, an access-brokering layer between agents and internal data looks like a durable architectural need.
Why CISOs Should Care
Lets a CISO permit broader internal AI agent adoption while keeping an identity- and policy-based checkpoint on what those agents can actually access.
What Makes It Different
An open-source identity- and policy-aware broker placed between agents and internal systems, rather than a closed point product bolted onto one agent platform.
The Matrix Verdict
63/100 — INCREMENTAL INNOVATOR
An Incremental Innovator: credible funding and a sensible open-source approach to a real problem, held back from a higher tier by unverified efficacy and adoption claims.
Editorial Note: Claims vs. Verified Findings
Claims of Fortune 500 production deployments and complete data-leak protection are company-stated and were not independently verified in this research.
Sources
Alternatives to Archestra.AI
Adaptive Security
AI-driven platform that simulates deepfake, voice, and multichannel social-engineering attacks to train and test organizations against next-generation phishing.
Quilr
CISO ReviewedEarly-stage agentic AI security startup building a 'Service-as-Software' platform to guard against human-related breaches and secure AI agent…
Zenity
Governance and security platform for AI agents and low-code/no-code development, securing agent identity, permissions and behavior across the…
Gray Swan AI
Carnegie Mellon-linked startup building automated red-teaming and runtime guardrails that frontier AI labs use to stress-test their models…
Tenzai
An agentic AI penetration testing startup building autonomous 'AI hackers' to find and validate exploitable vulnerabilities at a…
Reco
Reco secures the "agentic ecosystem" — mapping what AI agents can access across SaaS and enterprise apps, detecting…