Pomerium Agentic Access Gateway – dynamic auth for AI agents
TL;DR: We are building a new Agentic Access Gateway in Pomerium to safely let AI agents (like GPT-based deep researchers, scripts or assistants) access internal apps and resources on your behalf – with fine-grained, just-in-time authorization for every action. It's open source (GitHub link below) and we're looking for feedback and early access users. What is Pomerium? For those unfamiliar, Pomerium is an open-source identity-aware proxy (a "zero trust" access gateway). It sits in front of your internal apps and APIs, continually verifying identity and context on every request. The problem:…
In plain words
Pomerium Agentic Access Gateway is an open-source identity-aware proxy that controls how AI agents access internal applications and resources. It sits between agents and internal apps, applying fine-grained, just-in-time authorization to each action an agent attempts. The tool is designed for organizations deploying autonomous AI agents that need secure access to company data and systems while maintaining zero-trust security principles.
written from the facts on this page · September 2026
From the sources
In the maker’s words, at launch
TL;DR: We are building a new Agentic Access Gateway in Pomerium to safely let AI agents (like GPT-based deep researchers, scripts or assistants) access internal apps and resources on your behalf – with fine-grained, just-in-time authorization for every action. It's open source (GitHub link below) and we're looking for feedback and early access users. What is Pomerium? For those unfamiliar, Pomerium is an open-source identity-aware proxy (a "zero trust" access gateway). It sits in front of your internal apps and APIs, continually verifying identity and context on every request. The problem: AI agents are starting to act on our behalf in software – making requests, pulling data, and triggering actions autonomously. The rise of AI agents and protocols like Model Context Protocol (MCP) is really exciting. The potential for agents to interact with diverse tools (APIs, databases, SaaS) both internal and hosted to perform complex tasks is immense. However, the current MCP spec focuses on tool interaction and discovery but leaves per-request authorization largely undefined. Relying solely on initial OAuth scopes, as suggested, falls short for dynamic agent workflows where context can change mid-task. Pushing complex, context-aware AuthZ logic into every single tool creates security sprawl, inconsistency, and operational overhead – antithetical to core Zero Trust principles. Our solution: Agentic Access Gateway is a new feature in Pomerium designed for this AI-driven world. It extends Pomerium's core capabilities (continuous authn/authz) to non-human agents. In a nutshell, it treats AI agents as first-class identities that carry context and require policy checks at each step. Key functionality includes: - Centralized Policy Enforcement: Pomerium acts as a gateway in front of your MCP tools (and potentially other APIs agents might use). One place to define and enforce access policy. - Context-aware policy enforcement: Every request from the AI agent is checked against policy – including who (or what) the agent is acting for, what data it's trying to access, and any anomaly in behavior. If an agent strays out of bounds, it's denied on the spot. - Leverages Existing Identity: Agents authenticate via standard flows (OAuth2.1/OIDC style), so you can tie an agent's actions back to a real user or service account. Example: an agent acting for user Alice can inherit Alice's permissions (but only the ones you allow, and only while performing the task). - Just-in-time credentials: Instead of static API keys, an agent can request access through Pomerium and get a short-lived token scoped to the specific task or tool. No more "one token to rule them all" lying around. - Audit & traceability: All agent actions pass through a single gateway, so you get centralized logs and visibility. It's easy to see "which AI did what, when" for compliance or debugging. - Works with existing tools: Because it's built into Pomerium, you don't need a whole new stack. You configure policies in one place, and your internal APIs don't have to be modified. Demo: We made a 60s video showing Pomerium can protect access to both SaaS (Google Docs) and an internal apps (a internal db). See Claude pull data from a Google Doc, then pivot to an internal Postgres query – all in one run. https://www.youtube.com/shorts/IwMmuI-DMhs The Ask: We'd love the HN community's feedback on this approach. Are you dealing with AI agents in your systems yet? Sound interesting? Looking leverage an internal datasource to your LLMs? Sign up for early access to the Agentic Access Gateway: https://www.pomerium.com/secure-agentic-access If you'd like to contribute or want to dig into the code: https://github.com/pomerium/pomerium Thanks for reading! We built this because we believe the age of AI agents calls for a new kind of access control. Let us know what you think!
More ai this month
the category →
I trained a 125M-parameter transformer to autocomplete piano performances in real time (~108 notes/sec on an iPhone 15). The idea is basically GitHub Copilot or Tabnine, except instead of prompting it with code, you prompt it by playing a few notes on a MIDI piano. The model then continues what you played, entirely on-device. The app is free if anyone wants to try it. Happy to answer questions about the model, training, Core ML, or the many things that didn't work.
AI · 17d ago · simedw.com
Astute▲585Automate your B2B brand going viral, with new media creators
AI · 18d ago · company-app.joinastute.com
Hey HN, Henry from Cactus here! We previously released Cactus Needle, a 14MB agentic LLM for tool call, device use, and structured extraction for phones, wearables, smart homes, small robots and microcontrollers. We got really great feedback here, and have now incorporated the suggestions to release Needle 2. The whole model is a single 14MB binary that runs a full session in 28MB of RAM; 45m parameters at 2bit compression. Needle hits 500 tokens/sec decode speed on a Raspberry Pi 5, sits between 400-1,500 tokens/sec on VR devices like Meta Quest 3S and Apple Vision Pro, and ranges…
AI · 26d ago · cactuscompute.com
Launched alongside, May 2025
the whole month →
- C9
Life & fun · 2025 · felixrieseberg.github.io
