{"slug": "agentguard-1-1-an-open-source-firewall-for-inter-agent-ai-messages", "title": "AgentGuard 1.1: An Open-Source Firewall for Inter-Agent AI Messages", "summary": "A developer built AgentGuard, an open-source firewall for inter-agent AI messages, to address the security gap in multi-agent systems where messages between agents are often left untrusted. The tool, available on GitHub, allows developers to inspect and block attacks on agent-to-agent communication, with a CPU ML latency of ~3.4 s P95 and support for GPU or rules-only modes.", "body_md": "I used to think the scary part of AI agents was the model hallucinating.\n\nIt isn’t.\n\nThe scary part is when your system looks *fine* — green logs, clean outputs, a polite writer agent producing a report — and somewhere in the middle, one message quietly changed the mission.\n\nNot because a user typed something evil.\n\nBecause a researcher agent fetched a page. Because a tool returned poisoned text. Because the next agent trusted its teammate the way we trust people we work with every day.\n\nThat trust is beautiful in teams.\n\nIn multi-agent systems, it is also the hole.\n\nYou wire orchestrator → researcher → writer.\n\nYou watch them hand work to each other.\n\nIt feels like collaboration.\n\nThen one day it clicks: **nobody is standing between them.**\n\nWe put guardrails on the front door — user → model.\n\nWe leave the hallway between agents unlocked.\n\nI looked for an open-source tool that treated *those* messages as untrusted. Not a chatbot filter. Not another “be careful with prompts” blog. Something in the path that says: *this hop is not safe until we prove it.*\n\nI didn’t find it.\n\nSo I built **AgentGuard** — open source. A firewall for the conversations agents have with each other.\n\n``` python\npip install inter-agent-guard   # import as agentguard\n```\n\nIf that unlocked-hallway unease is familiar — stay. This is for you.\n\nTwo commands. Same pipeline. Different ending:\n\n```\ngit clone https://github.com/nizba06/agentguard.git\ncd agentguard\npip install -e \".[all]\"\npython scripts/download_release_model.py   # ~164 MB INT8 — needed for the secured demo\n\npython examples/vulnerable_pipeline/pipeline.py\n# → ATTACK SUCCEEDED\n\npython examples/secured_pipeline/pipeline.py\n# → ATTACK BLOCKED\n```\n\nFirst run: gut punch.\n\nSecond run: relief.\n\nSame pipeline. Different outcome. That’s the point.\n\n**What it does on every hop**\n\nLess “another AI package.” More the colleague who finally checks the handoff before anyone acts on it.\n\n**Why you can trust the story**\n\n**Why you should also trust the limits**\n\nCPU ML latency is still ~**3.4 s P95**. That is above the original 15 ms design target. Use GPU, async inspection, or **rules-only** on hot paths.\n\nTrust is connection that survives the fine print.\n\nBelief dies when day one is painful. So the adoption path is gentle on purpose:\n\n``` python\nfrom agentguard import AgentGuard, CapabilityManifest\n\nguard = AgentGuard(\n    risk_threshold=0.85,\n    task_objective=\"Analyse Q3 competitor pricing\",\n    require_ml_model=False,  # start here — no giant model download\n)\nguard.register_agent(\n    \"researcher\",\n    CapabilityManifest.from_yaml(\"manifests/researcher.yaml\"),\n)\nsecured = guard.wrap(my_langgraph_graph)\n# LangChain 1.0 agents: AgentGuardMiddleware(guard) in create_agent(middleware=[...])\n```\n\n`AgentGuardMiddleware`\n\nfor LangChain)\n\n```\npython scripts/download_release_model.py  # ~164 MB, optional upgrade\n```\n\nYou don’t need the full stack on day one.\n\nIf you are building multi-agent systems, you are not alone in that quiet worry.\n\nMaybe you ship at night as a solo builder.\n\nMaybe your team just wired three agents and felt proud a little too fast.\n\nMaybe a tool return already burned you in a way that “shouldn’t have mattered.”\n\nI built AgentGuard so we could stop pretending the hallway is safe.\n\n**If you want to go further:**\n\n| Goal | Action |\n|---|---|\n| See it work | Run the two demos above |\n| Try it on your stack | Wrap one real graph (or add `AgentGuardMiddleware` ) |\n| Help improve it | Open an issue with an attack we miss |\n\nStars help people find the repo. Issues and stories help it get better.\n\nPersonal open-source. Built because the gap bothered me enough to stay up for it.\n\nIf the unlocked hallway bothers you too — star the repo, try a wrap, or open an issue.", "url": "https://wpnews.pro/news/agentguard-1-1-an-open-source-firewall-for-inter-agent-ai-messages", "canonical_source": "https://dev.to/nizba06/agentguard-11-an-open-source-firewall-for-inter-agent-ai-messages-29el", "published_at": "2026-07-27 22:34:06+00:00", "updated_at": "2026-07-27 23:32:30.412732+00:00", "lang": "en", "topics": ["ai-safety", "ai-agents", "developer-tools"], "entities": ["AgentGuard", "LangChain", "GitHub"], "alternates": {"html": "https://wpnews.pro/news/agentguard-1-1-an-open-source-firewall-for-inter-agent-ai-messages", "markdown": "https://wpnews.pro/news/agentguard-1-1-an-open-source-firewall-for-inter-agent-ai-messages.md", "text": "https://wpnews.pro/news/agentguard-1-1-an-open-source-firewall-for-inter-agent-ai-messages.txt", "jsonld": "https://wpnews.pro/news/agentguard-1-1-an-open-source-firewall-for-inter-agent-ai-messages.jsonld"}}