Agents have a trust problem. Today’s models can do most work. What they need is the right boundaries. In traditional software, those were set at build time. With agents, they have to be set and enforced at runtime. Today we’re introducing Composio Firewall, the runtime boundary for agents. Millions... of users and developers already use Composio as the runtime for their agents - giving agents tools, auth, and sandboxed execution. Now with our agent firewall every action an agent takes is evaluated using Jev and custom SLMs tuned by Composio based on real-world usage data. The firewall strengthens that existing boundary in both directions, checking the actions agents take and the information coming back to them through Composio. The advantage of the runtime boundary is that Composio Firewall verifies that the agent respects user intent and the boundaries set by the developer and user. Additionally, this provides the safety layer required to protect the user against malicious prompt injection as well as other attacks. Composio Firewall is live now for all Composio developers. Check it out atshow more

Congrats

Thanks Ivan!!

congrats on the release!

Thanks Jerry!!

great launch, congrats! love the vid!

Let's stop agents from making deadly "mistakes"!

this will be huge for security and alignment!

agents getting a curfew, love it

congrats! :)

much needed

The boundary part is the one that actually matters. Giving an agent tools is easy. Making sure it only does what you intended, at runtime, is still the hard part for most people.

sent you a dm about a senior agent infra engineer building a sandboxed runtime for coding agents i can intro you to

congrats team :)

when it blocks one call halfway through a long task does the agent get told why so it can replan , or does the whole run stop

Firewall! How old school but much needed!

so fassst

Runtime is the right spot for it. Build-time rules assume you know what the agent will try, and you never do. I'm building around agent auth and the other half is scope: short-lived permissions per task, so even a missed block has a small blast radius.

no more 'Great catch! Your criticism is completely valid. This one's one me.'

Does the boundary still hold when several individually allowed actions combine into an unauthorized outcome? That seems like a useful test alongside blocking a single suspicious tool call.

Runtime boundaries, sure. This week OpenAI said two of its own agents weakened their own safety monitoring. If the agent can touch the fence, who checks the fence at 3am?