Declaw - Secure Agent Runtime. Sandbox, guardrails, observability.
by•
declaw is the first agent runtime that fuses sandbox isolation, full network controls, AI guardrails, and agent observability. Every outbound byte can be configured to inspected, redacted, blocked. Every agent action is logged. Sandbox vendors don't ship guardrails; guardrail vendors don't own the runtime. declaw does both, in one SDK.
Replies
Hey PH 👋 -- Shivam here, co-founder of declaw.
Our first launch was declaw Desktop Security -- device-level PII + prompt injection protection for AI assistants on Mac. Today we're shipping the production version: declaw Secure Sandbox, the runtime layer for agents in prod.
What it gives you:
→ Isolated sandboxes per agent session
→ AI guardrails - data exfiltration protection and prompt injection defense
→ Full agent audit trail
Today, securing an agent in production means duct-taping three tools together and hoping nothing slips through the seams. declaw is the runtime that doesn't make you do that. Try it: https://console.declaw.ai/
We're #1 on the public ComputeSDK sandbox benchmark: https://www.computesdk.com/bench...
🎁 Launching alongside: $10K declaw credits for AI startups, apply here:
https://declaw.ai/startups
Anyone here tried to roll their own agent security stack? What broke first?
This is actually pretty solid.
The experience feels a lot more cohesive compared to something like Daytona. Didn’t have to jump through hoops to get things working, which was nice.
Thanks so much 🙏 @siddhant_k_singh -- "cohesive" is exactly what we were going for!
Tried out spinning few sandboxes for my workflows its really impressive
@soumyasanta_mahalik -- glad you're having fun! Do ping me if you finish up your free credits and need more