The context window is a supply chain.
We've got the receipts.
opgate is built on mlld, our information-flow-control language designed for deterministically programming and securing LLMs. (Apache 2.0)
From the creators of npm audit
opgate receiptAgents need contextual guardrails.
“…instructions that are benign or even beneficial in some environments can be malicious in others.”
How do you lock down tools based on what the agent has accessed and what permissions the user has?
You can’t sandbox the world.
As agents increase in capability and access, they quickly outgrow sandboxes.
How do you lock down the semantic layer instead of just building a wider perimeter?
Everyone’s writing code now.
From the CEO to marketing and finance teams — most people vibe-coding aren’t developers.
How do you empower more of that? Securely? Cohesively?
Make the agent prove it.
CISO
Org-wide control over what agents are allowed to do based on what they’ve read.
CTO
All agent-written code follows our dev standards before a human even reviews the PR.
Firewall
Classify data and tune where it can flow based on your threat model. Every tool-call request is allowed, blocked, or escalated based on where the inputs came from.
Policy is a plain file in your repo, written in mlld. Classification happens as agents read — the classifier can only raise sensitivity; only a human lowers it; unclassified fails closed.
Receipts and sessions
Make agents show their work. Require coding agents to adhere to strict requirements for commits and PRs — repos declare checks; receipts prove the process ran.
Run code reviews using contributor inference — so maintainers stop having to pay for bugbot tokens.
Inspect, share, rehydrate, and interact with agent sessions — with log access governed by the same data policy as everything else.