generated: '2026-08-30'
method: searched
source: https://agentgateway.dev/docs/standalone/latest/quickstart/ + .../setup/ui/ + .../reference/cel/playground/ + .../llm/playground/
provider: AgentGateway
providerId: agentgateway
description: >-
  Agentgateway ships no hosted sandbox and issues no test credentials, because it is not a hosted
  service - there is nothing to sign up for. What it does ship is a set of SELF-HOSTED consoles and
  rehearsal surfaces that run inside the binary you install, plus quickstarts that stand up throwaway
  local backends. Everything below runs on the reader's own machine.
hosted_sandbox: false
test_credentials: false
test_credentials_note: >-
  No test keys, no test-mode prefixes, no test cards, no magic identifiers. Credentials in
  agentgateway are the reader's OWN upstream provider keys (for example $OPENAI_API_KEY), supplied
  through their configuration file. NEVER treat a key in the docs as a usable test value.
consoles:
  - name: agentgateway UI
    type: admin-console
    hosted: false
    docs: https://agentgateway.dev/docs/standalone/latest/setup/ui/launch-ui/
    detail: >-
      A web UI served on the port of any gateway listed in the `ui` section of the configuration. It
      manages the proxy, and since 1.5 renders logs and multi-turn agent trajectory views showing tool
      calls and results. Can be secured with a browser OIDC policy and exposed over HTTPS on your own
      hostname.
  - name: LLM playground
    type: playground
    hosted: false
    docs: https://agentgateway.dev/docs/standalone/latest/llm/playground/
    detail: Send a chat completion request through the gateway from inside the UI, without writing a client.
  - name: CEL playground
    type: playground
    hosted: false
    docs: https://agentgateway.dev/docs/standalone/latest/reference/cel/playground/
    detail: >-
      Evaluate a CEL expression in the UI before committing it to a policy. The nearest thing the
      product has to a dry run - see the reversibility and dry_run_mode blocks in
      conventions/agentgateway-conventions.yml.
  - name: CEL reference explorer
    type: interactive-reference
    hosted: true
    url: https://agentgateway.dev/docs/standalone/latest/reference/cel/cel-context-interactive/
    detail: An interactive browser of the CEL variable and function context, served on the docs site.
  - name: Configuration schema explorer
    type: interactive-reference
    hosted: true
    url: https://agentgateway.dev/docs/standalone/latest/reference/configuration/schema/
    detail: Generated from the agentgateway JSON schema, browsable in the docs.
quickstart_fixtures:
  note: >-
    The quickstarts stand up disposable local backends so a reader can exercise a route without any
    real upstream. These are third-party open-source test servers agentgateway points you at, not
    agentgateway-hosted fixtures.
  fixtures:
    - name: server-everything
      protocol: MCP
      docs: https://agentgateway.dev/docs/standalone/latest/quickstart/mcp/
      detail: An open-source MCP test server used as the MCP quickstart target.
    - name: httpbin
      protocol: HTTP
      docs: https://agentgateway.dev/docs/standalone/latest/quickstart/non-agentic-http/
      detail: A local httpbin used as the non-agentic HTTP quickstart backend.
    - name: A2A Hello World agent
      protocol: A2A
      docs: https://agentgateway.dev/docs/standalone/latest/agent/a2a/
      detail: >-
        The a2aproject/a2a-samples Hello World ADK agent on localhost:9999, proxied through a
        gateway on port 3000, with the sample CLI client verifying the agent card comes back.
    - name: examples/ directory
      url: https://github.com/agentgateway/agentgateway/tree/main/examples
      detail: >-
        ~25 runnable example configurations in the repository, each with a README and most with a
        docker-compose.yaml - llm-basic, llm-cost-routing, llm-prompt-guard, llm-semantic-routing,
        mcp-basic, mcp-authentication, mcp-authorization, mcp-multiplex, mcp-openapi, traffic-a2a,
        traffic-jwt-sign, traffic-oidc, traffic-spiffe, traffic-ratelimiting-global and more.
audit_modes:
  - name: Guardrail audit mode
    setting: 'webhook.action: audit'
    docs: https://agentgateway.dev/docs/standalone/latest/llm/prompt-guards/webhooks/
    detail: Runs a prompt guard without blocking - it records what it detects and forwards the content unchanged.
  - name: failOpen rate limiting
    setting: 'remoteRateLimit.failureMode: failOpen'
    docs: https://agentgateway.dev/docs/standalone/latest/configuration/resiliency/rate-limits/#failure-behavior
    detail: Not a test mode, but the documented way to stop a limiter outage from denying traffic while you evaluate.
maintainers:
  - FN: Kin Lane
    email: kin@apievangelist.com