Vijil
Trust platform for enterprise AI agents that discovers, verifies, defends and evolves agents through a continuous loop.
At a Glance
Vijil Dome, the control engine, under Apache-2.0; runs in your own process.
Engagement
Available On
Listed Oct 2026
About Vijil
Vijil, Inc. builds trust infrastructure for enterprise AI agents. Its platform is organized as four modules: Discover (identify), Diamond (verify), Dome (defend) and Darwin (evolve), which can be used individually or as one loop. Dome is described as open source, and a CLI and SDK are installable via pip.
What It Is
Vijil is an agent trust and security platform. It gives each agent an identity, tests its behavior, enforces limits on what it can access or disclose, and proposes improved versions of the agent within those limits. The company says the aim is agents that are verifiably reliable, secure and safe.
How the trust loop works
- Discover scans code, cloud and endpoints to build one inventory of agents, assigning each a SPIFFE identity.
- Diamond runs bespoke test suites and adversarial evaluations at scale, scored on tests the agent never trained on, and produces compliance-ready reports.
- Dome guards LLM and tool calls, carries data labels across handoffs, emits OTEL metrics, and mounts with two lines of code.
- Darwin searches agent variants within constraints, hardens against production failures and proposes evidence-backed pull requests, each re-verified by Diamond.
Vijil says each pass leaves evidence the next uses: Diamond failures become Darwin targets, and Dome blocks become Diamond probes.
Developer workflow
A git-style CLI covers the lifecycle (discover, register, evaluate, protect, monitor, adapt, evolve) with lower-level commands for agents, evaluations, harnesses, policies and scores. Inside an agent, a control decorator applies a policy file to inputs and outputs. A Claude Code plugin is also offered.
Vendor-published results
Vijil reports that in its own evaluation of eleven models, ten scored higher on their worst 10% of responses with Dome enabled, with a median gain of 17.6 points; one model showed no change.
Community Discussions
Be the first to start a conversation about Vijil
Share your experience with Vijil, ask questions, or help others learn from your insights.
Pricing
Open source (Vijil Dome)
Vijil Dome, the control engine, under Apache-2.0; runs in your own process.
- Vijil Dome under Apache-2.0
- pip install vijil-dome
- Guards on inputs, LLM calls, tool calls, outputs
- Discover installers and browser extension are open source
- Diamond and Darwin clients are free to install but not open source
Hosted — free
One agent on the hosted console at no cost; access currently opened by Vijil on request.
- One agent
- One baseline evaluation a week
- One Quick adaptive run a month
- Discovery at any scale
- One agent enforced with Dome
Hosted — Team
Priced per agent on the hosted console, monthly or annual; the amount is not published.
- Per-agent pricing, never per evaluation or token
- Each agent evaluated on every change
- Quick, Standard and Deep adaptive runs
- Each agent enforced with Dome
- Darwin proposals for each agent
In your network
Annual subscription for the platform in your VPC, datacenter or air-gapped network; quoted once, no agent count or token meter.
- Runs in VPC, on-premises or air-gapped
- Every agent, every evaluation
- No agent count, no token meter
- Decision traces kept in your network
Capabilities
Key Features
- Agent discovery and inventory across code, cloud and endpoints
- SPIFFE identity per agent
- Bespoke test suites and adversarial evaluations
- Compliance-ready reports
- Guardrails for LLM and tool calls
- Data labels that follow data across handoffs
- OTEL metrics
- Evolution of agent variants with evidence-backed PRs
- Git-style CLI and Python SDK
- Claude Code plugin