Doberman

Guardrails, security, and governance · DobermanCore · Open source project

An AI guard dog that sits at runtime to gate inputs, outputs, and tool calls, stopping unsafe or unintended actions before execution. github.com ↗

Facts

DescriptionDoberman sits on the execution path between an AI coding agent and its tools via an MCP proxy or host hooks, issuing PASS, AUTH, or BLOCK verdicts on every tool call before it runs. It is local-first and open source, guarantees fail-closed behavior and raise-only learning, and includes a turn gate for prompt injection, a policy-as-code file, decision audit logs, and a benchmark harness. It works with Claude Code, Codex, OpenClaw, Cursor, and any MCP-compatible agent.

DeploymentOpen sourceOn-premDesktop

PricingFree

Target buyersengineering

IntegrationsClaude CodeCodex CLIClaude DesktopCursorOpenClawMCP compatible agentsModel Context Protocol

Compliance claimsAudit logs

Features · 15

runtime tool call gatingpass/auth/block verdictsmcp proxyhost hooksfail-closed enforcementraise-only policy learningprompt injection turn gatepolicy as codeapproval prompts with 2fadecision audit logguardrail pluginsbenchmark attack-block harnesslocal-first dashboardstorage and secret leak protectionsstrictness tuning modes

Dossier · model-written summary

Doberman is an early-stage, open-source (Apache-2.0), local-first AI agent security layer built for developer and security-team use. Its core differentiation is sitting directly on the tool-execution path as an MCP proxy or host hook, meaning it can physically prevent a blocked action from ever running -- not merely flag it after the fact. However, the project is self-described as "Alpha," carries no named enterprise customers, publishes no compliance certifications, and several known capability gaps (e.g., no runtime egress broker) are documented openly on the repo. For a large enterprise, Doberman is a compelling technical concept that is architecturally sound but operationally immature: no SLA, no SOC2, no support tier, and no commercial pricing signal. The space it targets (agentic AI runtime enforcement) is real and growing, but the vendor itself is unproven at enterprise scale.

Sources: 10

Timeline · 4 events

2025-06-01NewsDoberman-Core project page published on SourceForge describing open-source runtime security layer for AI agents ↗deepdive

2025-06-01NewsDoberman listed on Glama.ai MCP server directory as an adaptive authorization and runtime guardrails tool ↗deepdive

2025-06-01NewsDoberman-Core repository published on GitHub under DobermanCore org (Apache-2.0, Alpha status) ↗deepdive

2025-06-01Launchtrydoberman.dev product site launched advertising open-source AI agent security with PASS/AUTH/BLOCK gating ↗deepdive

Same category · Guardrails, security, and governance

Hoop Akto