The Speedscale alternative — self-hosted, no account, no Kubernetes

Speedscale earned its reputation in Kubernetes traffic replay, and proxymock — its free, local record-then-replay CLI — is a genuinely good way to get a mock in front of your service in minutes. Its protocol breadth is wider than ours: gRPC, GraphQL, Kafka, AMQP, SQS/SNS, Postgres, MySQL, DynamoDB, S3. If you need to replay a database or a message queue today, that breadth is real and Gostly does not match it. Two things are worth knowing before you lean on the “free” label, though: proxymock’s mock engine is a closed proprietary binary (Speedscale describes it as a local Speedscale Enterprise mock server) — only the example snapshots are Apache 2.0 — and running it requires a Speedscale account, with a browser sign-in or API key on first use and no documented offline mode.

Gostly is built on a different starting assumption: the whole product should run inside your perimeter, not just a data plane. So Gostly puts a Rust proxy in front of your upstream and runs as a self-hosted stack — proxy, control plane, dashboard, and inference — via Docker Compose. In LEARN mode it records your real traffic; flip to MOCK and it replays it byte-for-byte, including HTTPS. There is no Kubernetes to operate and no kernel-privileged DaemonSet to install — the proxy is a userspace process you point your service at. The licensed build makes only opt-out license-validation and counters-only calls; your recorded traffic, your mocks, and any PII in them never leave your host.

The behavior that surprises teams most is fall-through. proxymock replays by signature matching, and on a miss it falls through to the real provider — which means a request your recordings didn’t cover can quietly hit production, an unwanted data-egress surprise in a test. Gostly never does this. On a miss the proxy returns a configurable status (404 by default) and stops there; the live upstream is never called. The match path runs a cascade — session-verbatim, then exact, then a resource store that turns a recorded POST into the right GET-by-id, then smart-swap — and only at the very edge does AI gap-fill come in, grounded by the traffic you already recorded, with no model anywhere in the request hot path.

Speedscale’s enterprise tier captures production traffic through an eBPF DaemonSet, which runs with node-kernel privileges, and its control plane is vendor-hosted SaaS — in the managed tier, DLP-filtered traffic is forwarded to Speedscale’s cloud; BYOC moves the data plane onto your infrastructure, though the control plane stays theirs. That is the right shape for load and chaos testing at Kubernetes scale, which is Speedscale’s center of gravity. It is the wrong shape if your constraint is that recorded traffic must not leave your network and you would rather not grant a vendor kernel-level capture on every node. Gostly captures at a userspace proxy you control, in a container you run.

And because Gostly is your own single-tenant deployment, the controls a security review asks for are built in: SSO via SAML or OIDC, role-based access control across four roles, and an append-only audit log on Team — all self-hosted in the container. Each deployment is single-tenant, so your data is isolated by running in your own perimeter, never shared with another customer’s. Pricing is published and flat, where Speedscale’s Pro and Enterprise tiers are contact-sales. The honest trade is protocol breadth: proxymock replays gRPC, queues, and databases today, and Gostly does not yet.