Twitxy Saved

authzed

Verified account This account is verified. Learn more

@authzed

Creators of SpiceDB: github.com/authzed/spicedb

Read how ChatGPT uses AuthZed for authorization: authzed.com/customers/open…
NYC / Remote authzed.com Joined July 2020
1.1K Following 1.6K Followers 855 Posts

Latest posts 5

authzed Verified account @authzed
We built our own internal Claude Design. It’s the first in our series on building agents on OAP.

Our version, messagingbot, lets us choose our model, control where our data goes, and fine-tune it to meet our requirements.

Claude Design is Anthropic’s tool for designing in conversation: describe a page, Claude builds it in your brand, and you refine it by commenting on the design.

messagingbot gives us the same but we didn’t have to write any platform code. messagingbot is built entirely from the agent primitives provided by OAP.
authzed Verified account @authzed
Giving an agent access to your data is easy. Making sure it only sees what the user is allowed to see is the hard part.

Our DevRel Sohan showed this at Open Source Summit, live-coding it with SpiceDB, Weaviate, and LangChain.

youtube.com/watch?v=tV4J4o…
www.youtube.com Building Fine-Grained Permissions for AI Agents - Sohan Maheshwar, AuthZed Join us at the premier vendor-neutral open source conference, where...
authzed Verified account @authzed
First in our series on agents built with Open Agent Primitives: a design agent for our marketing team, with no new platform or agent code.

Read-only access to approved brand repos, enforced outside the model. Every session leaves a signed audit log.

authzed.com/blog/open-agen…
authzed.com Open Agent Primitives: How We Build AI Agents We Actually Trust OAP is an open source platform for running enterprise AI agents, built on six primitives. See how AuthZed used it to build and secure a design agent.
authzed Verified account @authzed
We've been running AI agents at AuthZed that we actually trust. The platform behind them, Open Agent Primitives (OAP), is now open source.

We started where a lot of teams start: personal agents running with one person's credentials, hopefully scoped correctly, inside a sandbox that was hopefully configured correctly, and that was good enough to show us agents could be useful.

But we're an authorization company, and we knew there was a better way:

• Access controls could be more granular than whatever a given credential allows
• Actions could be permitted, rather than just contained by a sandbox
• An agent could act on behalf of someone or run independently with an identity of its own

• A team, not just an individual, could collaborate with an agent, and most importantly, the agent would stick to the permissions granted to that team

We distilled our observations and lessons into a set of primitives we think any enterprise needs to build and run agents it can trust with critical actions and data. Then we implemented them. Since every enterprise has its own infrastructure and requirements, components plug in through defined interfaces. You can swap in your own sandbox, LLM provider, memory system and other backends and the primitives still hold.

We've been running it internally for several months, building, deploying, and working with agents on it.

We're releasing two things:

1. The Agent Primitives: conceptual definitions you can use to evaluate any agent platform, including one you build yourself

2. Open Agent Primitives (OAP): our open source implementation. It ships with working defaults, so anyone can deploy it and start building agents within minutes (assuming you have a spare Kubernetes cluster handy). It includes example agents and tutorials to help you build your own.

We'll be following up with a series on how we built different types of agents on OAP. In the meantime, read the definitions, install OAP, and try it out.

• Read the definitions: authzed.com/ai/agent-primi…
• Try OAP: openap.org
• Source: github.com/authzed/openag…