Your code, in a world you control.

Build and test against realistic APIs you control. Reproduce problems, explore edge cases, and keep working when dependencies change.

Start free trial

Real traffic. A reusable starting point.

Capture the API behavior your application sees, redact sensitive headers before local disk capture, and turn recorded responses into a library you can replay.

01Capture the exchange
Keep the request and the response together.
02Redact sensitive headers
Authorization is scrubbed before disk capture by default.
03Build a replay library
Store behavior locally. Select MOCK to serve it.

Swipe to explore all five →

Version the environment alongside the code change.

Take one checkout integration from your laptop to the next release. Keep the dependency behavior you need, then change the conditions you want to test.

Local development

Keep building when the sandbox can’t keep up.

Record the payment provider’s responses and prepare a local replay. Work on checkout without making the provider part of every request.

Pull request

Run the same world in CI.

Bring the recorded payment behavior into CI, so an unavailable sandbox doesn’t have to interrupt your integration tests.

Staging and UAT

Reach the edge cases before your customers do.

Introduce a slow payment response or a rate limit. Advance virtual time in Time Chamber to reach token expiry without waiting an hour.

Production support

Understand what changed around a failed payment.

Compare fresh traffic with your recorded baseline. Inspect the difference, review a proposed mock update, and choose what to bring into the next test.

Your infrastructure. Your rules.

Run the proxy, dashboard, storage, and local AI in your own environment. Control what you capture, what you store, and how AI learns and responds.

Your infrastructureSelf-hosted

A stack under your control.

Your application

ProxyCapture + replay
Dashboard + APIConfigure + manage

Optional AIPro / Team

Inference + trainingRoute requests + train models
Local model serverServe model responses

PostgresRecordings + mocks

Setup onlySchema migrations

Prepare the database, then exit.

Your proxy, dashboard, database, and AI services run where you deploy them.

How this works

The proxy, control plane, Postgres, and local inference run in your deployment. Captured payloads are excluded from Gostly telemetry.

The dashboard and API share the ghost-web container. Pro and Team add ghost-inference for inference routing and training, plus ghost-llamacpp for model serving. The ghost-migrate container runs database migrations during setup and then exits.

Connected editions check licensing and can send opt-out usage counts separately from captured payloads.

Your infrastructureSelf-hosted

A stack under your control.

Your application

ProxyCapture + replay
Dashboard + APIConfigure + manage

Optional AIPro / Team

Inference + trainingRoute requests + train models
Local model serverServe model responses

PostgresRecordings + mocks

Setup onlySchema migrations

Prepare the database, then exit.

Your proxy, dashboard, database, and AI services run where you deploy them.

How this works

The proxy, control plane, Postgres, and local inference run in your deployment. Captured payloads are excluded from Gostly telemetry.

Connected editions check licensing and can send opt-out usage counts separately from captured payloads.

Let your coding agent set up Gostly.

For Claude Code, Codex, Cursor, or your preferred coding agent.

Copy setup prompt
Review proposed setup
Start Gostly locally

Setup prompt

Help me set up the licensed Gostly Docker Compose stack for this repository, then prove that one local API interaction can be recorded and replayed.

Read https://gostly.ai/llms.txt, https://gostly.ai/docs/install.md, https://gostly.ai/docs/quickstart.md, and https://gostly.ai/docs/record-and-replay.md first. Follow linked configuration, networking, or API documentation when needed. Use the current account-provided Compose and environment template as the installation artifacts. Do not invent a CLI, image registry, release tag, API call, or unsupported integration. If documentation and the downloaded artifacts disagree, identify the exact conflict and resolve it from the shipped configuration or API schema before proceeding.

Full setup prompt

Help me set up the licensed Gostly Docker Compose stack for this repository, then prove that one local API interaction can be recorded and replayed.

Read https://gostly.ai/llms.txt, https://gostly.ai/docs/install.md, https://gostly.ai/docs/quickstart.md, and https://gostly.ai/docs/record-and-replay.md first. Follow linked configuration, networking, or API documentation when needed. Use the current account-provided Compose and environment template as the installation artifacts. Do not invent a CLI, image registry, release tag, API call, or unsupported integration. If documentation and the downloaded artifacts disagree, identify the exact conflict and resolve it from the shipped configuration or API schema before proceeding.

1. Inspect and propose.
Check that the Docker daemon is reachable, Compose is installed, and Docker has enough allocated memory, disk, and the right CPU architecture for the selected images. Inspect this repository's runtime, development/test commands, CI, and HTTP dependencies without displaying secret values. Ask whether this is a fresh install, which Gostly plan I use, and where my downloaded installation files are. Identify existing containers, occupied ports, and persisted Gostly data before proposing changes.

Propose one small local test: app -> Gostly -> an explicitly approved non-production upstream, with the exact routes, file changes, ports, data directories, and test commands. Prefer a synthetic local upstream for the first proof. Explain how the app reaches Gostly and how Gostly reaches the upstream from their respective host/container networks: localhost inside a container is that container, not the host. Use Compose service names on a shared network, or the documented host gateway for the platform. Changing an app's base URL is different from configuring an HTTP/HTTPS forward proxy. Include service registration/routing where the selected workflow requires it. Leave CI changes as a follow-up plan. Wait for my approval before edits or traffic capture; honor any explicit approval already given for this exact scope.

2. Obtain the right installation artifacts.
For a fresh install, direct me to https://gostly.ai/register or https://gostly.ai/dashboard. I will handle account verification, plan selection, any billing confirmation, license issuance, and registry login myself. Do not sign me up or start a paid trial automatically. Ask me to download the plan-specific Compose and environment template into the chosen folder. Do not overwrite an existing .env. Only propose images actually named in these artifacts, and call out mutable tags without inventing a pinned version.

For an existing deployment, preserve its Compose project, volume paths, passwords, and current tenant identity. Follow the install guide's tenant-preservation procedure under the old configuration before any license or tenant change; never substitute a new organization ID for the tenant of existing data. Do not recreate or upgrade an existing deployment as part of a fresh-install smoke test.

3. Prepare configuration without exposing credentials.
Create or adapt placeholders and environment-variable references, then let me fill real credentials into a private, gitignored local .env or my secret manager. Never ask for keys in chat, echo them, embed them in a command argument, or put them in committed files. Registry login should accept the token through password-stdin or a credential helper, not an inline -p token. Keep .env, credential files, captured traffic, mock bodies, and license caches out of source control. Validate configuration without printing expanded secrets; for example, docker compose --env-file .env -f docker-compose.yml config --quiet.

Check that each required variable is actually referenced or passed into the intended container; merely adding a value to .env does not inject it. Confirm upstream routing, shared data paths, database/API authentication, and the stable tenant pin required by the downloaded stack. For the local trial, bind published ports to loopback and avoid existing port/container-name collisions. Derive topology, optional services, and resource needs from the actual Compose, not the plan name or an older example. A Free installation may include a control plane and database; do not reduce it to a standalone proxy. ENABLE_GENERATION=false alone does not prove that a separate model sidecar is disabled. Do not remove service dependencies or weaken auth/TLS to make startup pass. If HTTPS interception is needed, scope CA trust to the test client and ask before changing system trust.

For scripted control-plane checks, use the stack's service-key authentication while keeping browser password auth and Secure cookies intact. If setting the GHOST_API_KEY override, reference the same private .env value in BOTH ghost-web and ghost-proxy; setting it on only one side can break route synchronization and capture sealing. Coordinate recreation of both test services after changing that value so neither retains an older key. Verify authenticated route synchronization and that the intended service owns the captured request before building its library. Never print the key or disable authentication to resolve a 401.

4. Start the approved local stack.
Show the exact pull/start commands, working directory, Compose project/files, and services to be started. Ask for approval if startup was not included in my approved plan. After approval, pull and start only this installation. Check bounded readiness and relevant service health, then open the local operator dashboard if this plan includes it. The hosted account dashboard at gostly.ai/dashboard is separate; its agent-connected indicator is an additional licensed-activation check, not proof that replay works. A proxy-only plan has no local operator dashboard. If credentials, images, resources, or a required service are unavailable, report the specific blocker and retain the usable work.

5. Prove record -> build -> replay.
Use only the approved synthetic or non-production request. Confirm routing and LEARN readiness, send the request through Gostly, and verify the captured interaction. A healthy proxy can still reject capture with agent_not_ready/upstreams_not_loaded: check its declared control-plane connection and routing initialization, allow a bounded readiness wait, and report the blocker if it persists. Never disable that guard or count the rejected request as a recording. For a stack with the control plane, use the documented build/transition workflow to create the durable library and wait for it to finish before selecting MOCK. Confirm the mode API's request shape from the running schema or use the operator dashboard; do not guess between conflicting documentation examples. For an explicitly supported proxy-only distribution, use its documented recording/library and mode controls instead of calling an absent control plane; do not claim first-time capture works if its routing prerequisites cannot be met.

Repeat the same request through Gostly in MOCK. Check the expected status/body and evidence of which replay path answered. Prove that the test upstream received no additional request using its request count, or stop only that disposable upstream and repeat. An HTTP 200 or healthy container alone is insufficient. Distinguish an in-memory session replay from a durable library; if claiming replay survives restart, restart only the disposable test proxy and verify it again. Do not contact a production provider, globally switch an existing deployment's mode, or stop a shared upstream.

Finish with the actual local URLs, changed files, routes exercised, observed record/replay results, and what remains untested or manual. Include exact commands to stop only the test project while preserving its data; do not delete volumes. Clearly distinguish a complete fresh licensed install from a test using existing images/configuration or a documentation-only check.

For the licensed stack, create an account and issue a license key before startup.

Build in a world you control.

Capture real API behavior, replay it locally, and test what happens next.

Start free trial Book a demo

Your code, in a world you control.

Real traffic. A reusable starting point.

01Capture the exchange

02Redact sensitive headers

03Build a replay library

Version the environment alongside the code change.

Keep building when the sandbox can’t keep up.

Run the same world in CI.

Reach the edge cases before your customers do.

Understand what changed around a failed payment.

Your infrastructure. Your rules.

A stack under your control.

Redact at each storage boundary.

Credential headers

Body data Separate redaction path

The right access. The right data.

Sign in with your identity provider

Team roles

Tenant A Scoped access

Follow requests. Attribute actions.

Machine-level tracing

Insights

Operator-level tracing

Control how AI learns and responds.

Set confidence thresholds

Control fine-tuning

Review every repair

A stack under your control.

Let your coding agent set up Gostly.

Build in a world you control.

Your code, in a world you control.Your code, in a worldyou replayreplaytrainshapeaccelerateexplore.

Real traffic. A reusable starting point.

01Capture the exchange

02Redact sensitive headers

03Build a replay library

Version the environment alongside the code change.

Keep building when the sandbox can’t keep up.

Run the same world in CI.

Reach the edge cases before your customers do.

Understand what changed around a failed payment.

Your infrastructure. Your rules.

A stack under your control.

A stack under your control.

Build in a world you control.

Your code, in a world you control.