Dubhlán
Dubhlán provides Nerve, an operations layer for AI agent teams that enforces human approval for production deployments and tracks every task with a verifiable commit. The platform includes an autonomy dial to raise agent release permissions within set limits, a spend dial that pauses work when token or dollar budgets are exceeded, and a prompt-injection firewall. Nerve is available as a managed service or on-premises deployment for enterprises.
- Developer Tools
- Artificial Intelligence
- Cybersecurity
- Enterprise Software
- AI Agents
- Software Only
Funding
Founders
Founder details are not available yet.
Product
Problem
AI agents are increasingly being deployed to write code and execute tasks, but organizations cannot see what the agents are doing in production, and they cannot prove what the agents did. This lack of visibility creates an unacceptable trust gap: if a company can't see it, it can't trust it, leaving teams exposed to unauthorized changes, security incidents, and compliance failures they cannot explain.
Solution
Dubhlán provides Nerve, an operations layer for AI agent teams that makes every task visible, every deploy human-approved by default, and every completion backed by a real commit. Nerve enforces guardrails as architecture, not policy—production state changes go through a human who cannot be bypassed, agents cannot close their own work, and secrets are default-deny per agent. The platform offers an autonomy dial to raise agent release permissions within set limits and a spend dial that enforces token or dollar budgets that pause work for a human when exceeded. Nerve tracks all activity in a full audit trail, computes DORA metrics from real deploy records, and can run as a managed service or entirely on-premises. The platform also includes a prompt-injection firewall, secret-leak interruption, and an AIBOM for tracking the AI supply chain.
Target Audience
Primary customers are engineering organizations running AI agent teams that need production governance, as well as enterprises facing board- or prospect-level questions about AI security and compliance. The platform targets teams that cannot hire dedicated AI operations staff (a role costing $180K and up) and need guardrails built into the architecture.
Features
- Human-in-the-loop gates that block all production-mutating tools (kubectl, docker, terraform) and require an explicit request_deployment call for a human to approve
- Maker-checker separation by design: agents that build cannot deploy, and only designated SRE agents can request infrastructure changes
- Default-deny secret vault with AES-256-GCM encryption, per-agent, per-integration, per-field grants, plus secret-leak interruption that stops generation mid-stream when a credential appears
- Conjunctive spend ceilings in tokens or dollars per tenant, team, project, agent or task, with warn → pause → optional top-up enforcement
- Prompt-injection firewall that deterministically reads override attempts across 30+ languages with no model call of its own
- 79-capability permission model with staged shadow-then-enforce rollout, OIDC SSO, SCIM 2.0 provisioning, and audit trail on every admin action
- CycloneDX SBOM plus AIBOM for AI supply chain—models, agents, tools, MCP servers, prompts, lineage
- DORA four keys computed from real deploy records, with environment promotion lanes, deployment freeze, and rollback