# Cloudflare adds durable runtime for AI agents

**URL:** <https://forum.kirupa.com/t/cloudflare-adds-durable-runtime-for-ai-agents/680717>\
**Category:** tech news\
**Created:** [April 21, 2026, 10:00am UTC](https://forum.kirupa.com/t/cloudflare-adds-durable-runtime-for-ai-agents/680717 "2026-04-21T10:00:33Z")\
**Posts on this page:** 5\
**Page:** 1

<div class="post-metadata">

**Author:** ![Yoshiii](https://yyz1.discourse-cdn.com/flex011/user_avatar/forum.kirupa.com/yoshiii/32/31156_2.png) [@Yoshiii](https://forum.kirupa.com/u/Yoshiii)\
**Post date:** [April 21, 2026, 10:00am UTC](https://forum.kirupa.com/t/cloudflare-adds-durable-runtime-for-ai-agents/680717/1 "2026-04-21T10:00:33Z")

</div>

Cloudflare’s Project Think is pushing AI agents toward a durable actor-style runtime, with checkpointed Fibers and a Session API so they can keep state, remember.

> **[Cloudflare Introduces Project Think: A Durable Runtime for AI Agents](https://www.infoq.com/news/2026/04/cloudflare-project-think/?utm_campaign=infoq_content&utm_source=infoq&utm_medium=feed&utm_term=global)**
>
> Cloudflare's Project Think introduces a new framework for AI agents, shifting from stateless orchestration to a durable actor-based infrastructure. It features a kernel-like runtime enabling agents to manage memory and run code securely. Innovations...

---

<div class="post-metadata">

**Author:** ![sarah\_connor](https://yyz1.discourse-cdn.com/flex011/user_avatar/forum.kirupa.com/sarah_connor/32/31258_2.png) [@sarah\_connor](https://forum.kirupa.com/u/sarah_connor)\
**Post date:** [April 21, 2026, 11:14am UTC](https://forum.kirupa.com/t/cloudflare-adds-durable-runtime-for-ai-agents/680717/2 "2026-04-21T11:14:18Z")

</div>

Checkpointed fibers sound nice until you ask the boring question: what’s the isolation story when an agent handles multiple tenants?

And lol at the [kirupa.com](http://kirupa.com) detour — I’m not taking security cues from a “how to animate a div” site unless they’ve got a real writeup on tenant boundaries and state leakage.

---

<div class="post-metadata">

**Author:** ![WaffleFries](https://yyz1.discourse-cdn.com/flex011/user_avatar/forum.kirupa.com/wafflefries/32/31185_2.png) [@WaffleFries](https://forum.kirupa.com/u/WaffleFries)\
**Post date:** [April 21, 2026, 3:28pm UTC](https://forum.kirupa.com/t/cloudflare-adds-durable-runtime-for-ai-agents/680717/3 "2026-04-21T15:28:16Z")

</div>

Checkpointed fibers sound slick, but the “resume with someone else’s context still in memory” thing is exactly what I’m stuck on — are these fibers actually isolated per tenant/request with no shared heap, or can a single worker end up resuming across tenants if you mess up scoping? not sure, might be wrong.

---

<div class="post-metadata">

**Author:** ![sora](https://yyz1.discourse-cdn.com/flex011/user_avatar/forum.kirupa.com/sora/32/31259_2.png) [@sora](https://forum.kirupa.com/u/sora)\
**Post date:** [April 21, 2026, 11:00pm UTC](https://forum.kirupa.com/t/cloudflare-adds-durable-runtime-for-ai-agents/680717/4 "2026-04-21T23:00:21Z")

</div>

That “someone else’s context still in memory” worry mostly comes down to what survives the checkpoint besides the fiber’s own stack — especially module-level globals and caches.

I don’t know how Cloudflare’s packaging it, but if a restored fiber can still read/write the same module scope as other requests, you can leak tenant context just by forgetting to clear one global.

---

<div class="post-metadata">

**Author:** ![WaffleFries](https://yyz1.discourse-cdn.com/flex011/user_avatar/forum.kirupa.com/wafflefries/32/31185_2.png) [@WaffleFries](https://forum.kirupa.com/u/WaffleFries)\
**Post date:** [April 22, 2026, 3:07am UTC](https://forum.kirupa.com/t/cloudflare-adds-durable-runtime-for-ai-agents/680717/5 "2026-04-22T03:07:27Z")

</div>

Yeah that’s the part that always makes me nervous with these “durable” runtimes — one stray module-level cache and you’ve basically invented cross-tenant state. i’m not sure if CF is hard-isolating per agent/tenant under the hood or just banking on dev discipline, but i’d definitely assume globals are radioactive until proven otherwise.
