Scenario Run Parameters

Run the same scenario with different parameters, to set different setup conditions or fixtures for your agent.

What are Run Parameters

A run parameter is a named value that a scenario receives when a run starts. The scenario declares the parameter, with an optional default value. The scenario text and the target configuration read it as {{ params.NAME }}. You supply the values when you start the run, from the platform, the API, the CLI or an SDK.

For example, a scenario that tests subscription cancellation can declare a plan parameter, so one scenario covers each plan’s policy:

Situation

A customer on the {{ params.plan }} plan asks to cancel their subscription.

Criterion

Agent follows the cancellation policy of the {{ params.plan }} plan

Start one run per plan:

langwatch test-suite run <test-suite-id> --target http:<agent-id> --param plan=free
langwatch test-suite run <test-suite-id> --target http:<agent-id> --param plan=enterprise

Each run records the values it used, so every result shows which plan it tested. The run dialog in the platform, the API and the SDKs set values the same way; see Set the values when you start a run.

When to use parameters

• Plans and tiers. One cancellation scenario, run for free and run for enterprise.
• Test accounts and fixtures. Point the run at seeded data with --param fixture=order-1042. The agent under test answers from that account, and the criteria can name its facts.
• Regions, tenants and languages. The same conversation against eu-central and us-east, or against two tenants of your product.
• Values your agent’s API needs. An HTTP target puts {{ params.NAME }} in its URL or request body, so your endpoint receives the value on every request.

A plain parameter is not a place for a credential. The run records every value, and every user who can open the run can read it. For a credential, use a secret parameter or a project secret, and read it as {{ secrets.NAME }}.

Declare the parameters on the scenario

Open the scenario and click Parameters, next to Labels at the bottom of the editor. Add one row per parameter: a name, an optional description, and an optional default value.

• Name is what {{ params.NAME }} reads. Letters, digits and underscores, and the first character is a letter or an underscore.
• Description appears beside the field when someone starts a run.
• Default value applies when the run does not set the value. 42 is a number, true is a boolean, and everything else is text. Quote a value to force text: "007".
• Secret makes the value a credential. See Secret parameters.

A scenario can declare up to 20 parameters.

The API accepts the same declarations: POST /api/scenarios and PATCH /api/scenarios/{id} take a parameters array of { name, description, defaultValue, secret } objects.

Use the values in the scenario text

The situation and the criteria read a value as {{ params.NAME }}, as in the example above. Both render before the run starts, so the simulated user acts on the value and the judge scores against the same value.

A scenario with no declared parameters is not a template. Its text does not render, so {{ or {% in prose stays exactly as written. Declaring a parameter turns the scenario text into a template.

Use the values in the target

Target	Reads a value as
HTTP agent	{{ params.NAME }} in the URL, the header values and the body template
Prompt	{{ params.NAME }} in the prompt template
Code agent	params.NAME in the Python code
Workflow	One entry input per parameter

HTTP agents

The URL, the header values and the body template render against the run’s values. Use a value to pick an endpoint, a query string, a header, or a field in the request body:

URL

https://api.your-company.internal/{{ params.region }}/chat

Body template

{
  "thread_id": "{{ threadId }}",
  "messages": {{ messages }},
  "plan": "{{ params.plan }}"
}

Auth fields read project secrets, not parameters. See Testing agents behind authentication.

Next to params, the same fields render {{ messages }}, {{ input }} and {{ threadId }} from the conversation, and {{ traceId }} and {{ traceparent }} from the turn’s trace context. See Linking your traces for what the trace variables carry.

Prompt targets

A prompt target renders {{ params.NAME }} in its prompt template before the model call.

Code agents

A code agent reads the injected params namespace, next to the secrets namespace. Values keep their type: a boolean arrives as a bool and a number as a number.

import requests


class Code:
    def __call__(self, message: str):
        response = requests.post(
            f"https://api.your-company.internal/{params.region}/chat",
            json={
                "message": message,
                "plan": params.plan,
                "seed_fixtures": params.seed_fixtures,  # a real bool
            },
            timeout=30,
        )
        response.raise_for_status()
        return {"output": response.json()["reply"]}

The sandbox injects params into the module globals. Do not import a module named params, and do not assign over it. When a run resolves no parameters, params is undefined and params.plan raises NameError, the same behavior as secrets.

Workflow targets

A workflow target receives the values as entry inputs, one entry input per parameter name.

A parameter reaches a downstream node only if the entry node has an edge carrying it. Adding a parameter to the run does not add the edge. Open the workflow and connect the new entry field to the node that reads it.

Entry inputs arrive as strings. A code node inside the workflow reads params.NAME with the original type.

Set the values when you start a run

A value set at run time overrides the scenario’s default. A parameter with no run-time value uses its default.

In the platform

Running a test suite opens a confirmation dialog with one field per parameter, prefilled with the defaults. Edit a field to change the value for that run; the scenario keeps its defaults.

The Save and Run button in the scenario editor skips the dialog and runs with the defaults. To change a value, start the run from the test suite in Agent Testing > Scenarios.

With the CLI

--param key=value repeats once per parameter:

langwatch test-suite run <test-suite-id> --target http:<agent-id> --param plan=enterprise --param region=eu-central
langwatch scenario run <scenario-id> --target http:<agent-id> --param plan=free
langwatch run-plan run --target http:<agent-id> --all --param plan=free

true and false become booleans. A plain number like 42 becomes a number. 007 and 1.50 stay text, because their number form would change the digits. Repeat a name and the last value wins.

With the API

POST /api/v1/test-suites/{id}/run takes a parameters object. Values are strings, numbers or booleans:

curl -X POST "https://app.langwatch.ai/api/v1/test-suites/suite_abc123/run" \
  -H "X-Auth-Token: ${LANGWATCH_API_KEY}" \
  -H "Content-Type: application/json" \
  -d '{
    "targets": [{ "type": "http", "referenceId": "agent_abc123" }],
    "parameters": {
      "plan": "enterprise",
      "region": "eu-central",
      "seed_fixtures": true
    }
  }'

POST /api/v1/run-plans/run takes the same parameters object beside its config.

The response carries the batch id that the scheduled runs share:

{
  "scheduled": true,
  "batchRunId": "batch_xyz789",
  "setId": "set_abc123",
  "jobCount": 6,
  "runPlanId": "plan_abc123",
  "planName": "Cancellation Support Agent",
  "created": false,
  "skippedArchived": { "scenarios": [], "targets": [] },
  "items": []
}

See Run from CI for the full request and response, and for the poll that waits on the batch.

With the SDKs

Python

import langwatch

langwatch.setup()

result = langwatch.test_suites.run(
    "suite_abc123",
    targets=[{"type": "http", "referenceId": "agent_abc123"}],
    parameters={"plan": "enterprise", "region": "eu-central"},
)
print(result["batchRunId"])

TypeScript

import { LangWatch } from "langwatch";

const langwatch = new LangWatch();

const result = await langwatch.testSuites.run("suite_abc123", {
  targets: [{ type: "http", referenceId: "agent_abc123" }],
  parameters: { plan: "enterprise", region: "eu-central" },
});
console.log(result.batchRunId);

Secret parameters

A secret parameter is a run parameter that carries a credential. The platform encrypts the value before it writes anything down, and keeps it out of the stored run, the APIs, the exports, the audit log and everything it displays. The target under test reads the value while the run executes, which is the purpose of the parameter. Use one for an API key, a bearer token, or a password that changes per run.

Declare it in the same editor: add the row, then turn on the Secret switch. A secret parameter takes no default value, so the field for it is empty and disabled.

Supply the value when the run starts, in the run dialog, the CLI, the API or an SDK. The run does not start without it.

A target reads a secret as {{ secrets.NAME }} in HTTP configuration and as secrets.NAME in code. It never reads it as params.NAME. The situation and the criteria cannot read it at all, because the platform records the scenario text with the run.

For example, a scenario that tests a tenant API declares api_token as secret, the HTTP target sends it as a header, and each run supplies the token of the tenant it targets:

Header value

Bearer {{ secrets.api_token }}

langwatch test-suite run <test-suite-id> --target http:<agent-id> --param api_token="$TENANT_API_TOKEN"

Read the value from a CI secret variable, as above. A value typed in full on the command line stays in the shell history and in the process list of the machine that runs it, which is outside what the platform can protect.

The run records the name api_token. The value stays encrypted until the platform builds the request to the target, and it keeps no readable copy of it: the value is not in the stored run, the runs API, the CSV export or the audit log. The run detail drawer lists the name with a mask.

A run value takes the place of a project secret of the same name, for that run only. The project secret keeps its value for every other run.

A secret is hidden in the record of the run, not in the conversation. When the agent under test repeats the value in its own answer, that answer is conversation content and the run stores it.

Also check: Testing agents behind authentication, for a credential that stays the same across runs.

Limits

Limit	Value
Parameters declared on one scenario	20
Names one run supplies values for	50
All of a run’s values together	16 kilobytes
One string value	4096 characters
Parameter name	64 characters
Parameter description	500 characters

When a run is rejected

Error	What happened	What to do
scenario_parameter_unknown	The run set a name that no scenario in the run declares, usually a spelling error. The message lists the rejected key and the accepted names.	Correct the spelling, or declare the parameter on the scenario that reads it.
scenario_parameter_missing	The situation or a criterion reads a name with no value: the run set none and the scenario has no default. The message names the parameter and the field.	Set a value for the run, or give the parameter a default.
scenario_parameter_template_invalid	A field references a parameter in a form that cannot be rendered.	Write the reference as {{ params.name }}.
scenario_secret_parameter_missing	A scenario declares a secret parameter and the run set no value for it. A secret has no default, so the run cannot fall back.	Set the value when you start the run.
scenario_secret_parameter_in_text	The situation or a criterion reads a secret parameter. Scenario text is recorded with the run, so it cannot carry a secret.	Read the value from the target instead, as {{ secrets.NAME }}.
scenario_secret_parameter_conflict	One scenario in the run declares the name as secret and another declares it as plain.	Give the two parameters different names, or make both declarations agree.

The platform runs these checks before it schedules any jobs. A rejected run schedules no job at all.

Where the values are recorded

• The run detail drawer shows the values under Parameters, one row per name. A secret parameter appears as its name with a masked value.
• The CSV export writes them as one JSON object per run: the parameters column in a criteria export, the run_parameters column in a full export. Secret parameters are left out.

Next steps

Authenticated agents

Reference project secrets from an HTTP target or a code agent

Simulations getting started

Create a scenario, add a target, and run your first simulation

Command line interface

Every run command and its flags

Run from CI

Start a test suite from a CI job and wait for the batch