3 Cloudflare Workers alternatives for real-time agents

Cloudflare Workers is the default answer to where edge code runs. Near-zero cold starts, a network spanning 330+ cities, and a mature developer ecosystem got it there, and for JavaScript-first web workloads it remains a strong choice.

The isolate model that makes those cold starts fast is also what sets the ceiling. Workers supports JavaScript, TypeScript, Python, Rust, and WebAssembly, not Go or Java, and there is no way to run a container. KV, Durable Objects, and R2 cover storage and state, but there is no filesystem, no telephony layer, and Workers AI runs inference on GPUs Cloudflare rents rather than owns.

What actually decides whether a platform can carry a real-time or comms-heavy workload is whether it runs more than JavaScript, whether reaching storage and inference costs a network hop, and whether the platform can place a phone call at all. We ranked three providers on exactly that.

Here's three Cloudflare Workers alternatives for Functions, ordered from owned infrastructure outward.

Cloudflare Workers alternatives

1. Telnyx

• Best for: Voice AI agents, IVR replacement, and backend workloads needing storage, state, and telephony on one platform
• Deployment model: Containers on owned edge PoPs, deployed with telnyx-edge ship, with bindings to KV, StatefulActor, CloudFS, and Inference
• Pricing model: Usage-based, no egress between platform components

Any language, any runtime

Telnyx's Functions runtime runs containers, not isolates. TypeScript, Go, Python, and Java all deploy the same way, with one command: telnyx-edge ship. The platform builds the container, deploys it to edge PoPs, and returns a public URL.

The container is the entry point to a wider platform. A function reaches KV, StatefulActor, Cloud Storage, CloudFS, and Inference through bindings instead of network calls, so there are no credentials to manage and no request leaves the platform. On the Frankenstack most teams build by default, a call lands on Twilio, audio moves to Deepgram, text moves to OpenAI, a response comes back from ElevenLabs, and audio returns through Twilio to the caller: four hops, four vendors, four sets of credentials. On Telnyx, the call lands, a function reads KV, calls inference, and responds, all in the same facility, zero hops.

Running on carrier infrastructure

Telnyx is a licensed carrier and signs STIR/SHAKEN attestation as the originating carrier, the highest trust level the phone network supports. Cloudflare has no telephony layer, so it cannot participate in call attestation at all.

That carrier network is also where the edge PoPs sit. Telnyx runs 18 PoPs across the Americas, Europe, MENA, and APAC, and inference stays in-region as a property of the architecture, not a configuration option. A call into an EU PoP keeps its transcription, LLM routing, and speech synthesis in that facility from start to finish, and the audio never leaves the EU.

A function calls env.TELNYX.calls.dial(number) to place a call or env.TELNYX.messages.send(to, body) to send a message, with no separate telephony vendor and no API key in the repository. Five vendors stitched together at 99.9% uptime each compound to roughly 99.5% availability, close to 4.4 hours of downtime a month. A single 99.999% SLA works out to about 43 minutes.

2. AWS: Lambda, DynamoDB, S3, and Bedrock

• Best for: Teams already deep in AWS who can absorb the integration work
• Deployment model: Regional Lambda wired by hand to DynamoDB, S3, and Bedrock, with no edge-native path
• Pricing model: Per-invocation plus per-service billing, with egress fees between components

If the reason to leave Cloudflare Workers is needing more than an isolate can run, AWS looks like the obvious answer. Lambda handles compute, DynamoDB handles state, S3 handles storage, and Bedrock handles inference, and every one of those services is mature and battle-tested at scale.

None of it is edge-native. Lambda@Edge cannot attach to a VPC, so private networking runs through a separate path, and every hop between Lambda, DynamoDB, S3, and Bedrock carries its own IAM role, its own SDK configuration, and often an egress fee. There is no telephony colocation and no path to placing a call without adding a fifth vendor. Teams that want one platform instead of four services wired together, with bindings instead of IAM roles, tend to land on Telnyx.

3. Vercel Edge Functions

• Best for: Next.js and frontend-first teams
• Deployment model: Regional functions tied to a frontend deploy pipeline
• Pricing model: Usage-based, bundled with hosting plans

Vercel is a frontend platform first. Next.js integration, preview deployments, and edge middleware make it the fastest path from a git push to a live page, and Fluid Compute has cut cold starts further for Node.js workloads.

Edge Functions run in cloud regions, not at a true edge, and there is no platform storage, state, or inference behind them. A team building a chat widget on Vercel still needs a separate database, a separate model API, and a separate phone system to call anyone. Teams that want storage, state, and inference built into the compute layer, not bolted on through a third service, tend to land on Telnyx.

How we evaluated these edge compute platforms

We scored every platform on the four things that decide whether it can carry a real-time or comms-heavy workload. Cold start benchmarks were not one of them.

• Runtime flexibility: Whether the platform runs containers and any language, or restricts code to a JavaScript-compatible isolate.
• Binding model: Whether storage, state, and inference are reachable through a binding with zero hops, or require a network call to an external service.
• Owned infrastructure: Whether the platform runs on GPUs and PoPs it owns, or rents cloud capacity that adds a margin and a hop.
• Telephony and comms primitives: Whether the platform can place a call or send a message natively, or has no telephony layer at all.

Telnyx is the only platform here that clears all four, which is why it sits at the top of the list.

Choose the platform built for containers and comms

Cloudflare Workers wins on cold starts and ecosystem maturity, and for a JavaScript-only web workload that is often enough. Telnyx wins the moment a workload needs a container, a phone number, or a model running on infrastructure it owns.

Platform engineers get containers and real storage and state without wiring five services together by hand. Comms and telecom developers get an edge platform built on the carrier network that terminates the call, something no CDN-based platform can offer. Agent builders and ISVs building their own agent platforms get all of it on one bill, with bindings instead of credentials between the pieces.

Zero hops is the architecture.

Pick the alternative that matches what your workload actually needs:

• Cloudflare D1 alternatives when the state layer, not the compute layer, is the real constraint.
• Top edge compute platforms for a broader survey before narrowing to one.
• Edge functions explained for the fundamentals of what a function can and cannot do at the edge.
• Serverless functions explained when the comparison starts further back, at serverless versus edge itself.

The shortest path to the caller is the only latency budget that compounds.