PipeLab
PipeLab builds open-source agent-security tools, led by Pipelock, an agent firewall that sits between AI agents and the internet. Its focus is detecting and enforcing against credential leaks, prompt injection, SSRF, tool poisoning, and unsafe MCP/HTTP/WebSocket traffic while producing evidence that can be verified offline.
At a Glance
- Startups, small teams, and solo builders running multiple AI-agent profiles
- Enterprise security, platform, and compliance teams operating fleets of AI agents
- Organizations using MCP servers, coding agents, operations agents, and tool-enabled workflows
- AI-agent security researchers and developers evaluating egress controls
AI Tools by PipeLab
(1)Pipelock
Open Source AI Agent Firewall
Discussions
No discussions yet
Be the first to start a discussion about PipeLab
Latest News
Pipelock v3.6: Watch and Check Contained Agents; adds private network namespaces, agent viewing/control, signed workspace-change statements, provider-aware credentials, and fixes for two security advisories.
Pipelock v3.5: Native Agent Evidence Level Records, making firewall receipts consumable by the AEL reference checker.
Pipelock v3.7: Parallel Receipts and Filesystem Grants, with complete-group verification, filesystem grants, local MCP identity checks, and narrower policy grants.
Pipelock Benchmark Scores: PipeLab published a public scoreboard for the Agent Egress Bench corpus and reproducible tool-security results.
Products & Services
Open-source Go single-binary agent firewall for MCP, HTTP, and WebSocket traffic. It mediates traffic, scans content and tool calls, enforces DLP and egress policy, supports containment, and emits signed action receipts. Apache-2.0 core; enterprise features are under ELv2.
Paid offline-licensed coordination tier adding unlimited named security profiles, separate policies and budgets, per-agent identity and allowlists, operator dashboard, behavioral baselines, and profile-scoped audit trails.
Fleet control plane for signed policy distribution, fleet audit aggregation, agent enrollment and identity, rollback, and multi-party emergency kill controls.
Assessment capability and annual license that collects deployment evidence and produces signed HTML and JSON assessment output, remediation guidance, and framework mappings.
Market Position
PipeLab positions Pipelock as an open-source network/content enforcement layer rather than a model guardrail: it inspects what an agent actually sends and receives at the boundary, while producing independently verifiable signed evidence. Its comparison material places it against egress proxies and gateways such as agentgateway, AWS AgentCore Gateway, Envoy AI Gateway, Cloudflare AI Gateway, Docker MCP Gateway, and iron-proxy; against sandboxes/scanners such as ToolHive, Snyk Agent Scan, DefenseClaw, and NemoClaw; against guardrails such as Lakera Guard and LlamaFirewall; and against broader AI-security platforms such as Backslash, Prisma AIRS, and Runlayer.
Leadership
Founders
Joshua Waldrep
Master plumber by trade with 14 years of plumbing experience; built Pipelock from a homelab automation project and maintains the open-source project.
Executive Team
Joshua Waldrep
Founder and maintainer
Master plumber and open-source developer; founder of PipeLab and creator/maintainer of Pipelock.
Founding Story
PipeLab began as a 2024 homelab project for a plumbing business that wanted AI agents to handle dispatch, scheduling, and back-office work. The agents had shell access, API keys, and unrestricted network access, so Joshua Waldrep built Pipelock as the missing enforcement point between agents and the internet; the project grew from that operational need into a public security tool.
Business Model
Revenue Model
Open-source Apache-2.0 security engine is free. Revenue comes from offline-licensed multi-agent coordination (Pro), enterprise fleet governance via Conductor, and the paid Assess assessment license; there is no per-agent or per-request metering in the published plans.
Pricing Tiers
Forever free; full security engine with one default security profile.
Founding price for multi-agent coordination, named profiles, separate policies, budgets, audit trails, and operator features; the pricing page states the regular Pro price is $99/month.
Credited toward year one if the customer signs within 30 days; includes a signed Fleet Receipt Report.
Limited design-partner offer; renews at the published annual price unless otherwise agreed.
Flat fleet-governance contract with no seat or agent-count meter.
Adds full assessment report, server-specific findings, remediation guidance, detailed framework mappings, and optional offline-verifiable attestation and badge.
Target Markets
- Startups, small teams, and solo builders running multiple AI-agent profiles
- Enterprise security, platform, and compliance teams operating fleets of AI agents
- Organizations using MCP servers, coding agents, operations agents, and tool-enabled workflows
- AI-agent security researchers and developers evaluating egress controls
- Protecting AI agents and MCP clients from credential exfiltration and malicious or poisoned tool responses
- Controlling agent egress to APIs, websites, cloud metadata endpoints, private networks, and approved destinations
- Containing coding and operations agents with shell access, credentials, and workspace permissions
- Running multi-agent deployments with separate policies, budgets, identities, and audit trails
- Enterprise fleet governance, signed policy rollout, emergency kill, and offline-verifiable compliance evidence
- Benchmarking agent-egress security tools with reproducible public attack cases