Testing Locally

Before publishing an agent, you’ll want to test it thoroughly. Selu provides a local development workflow that mirrors the production environment as closely as possible.

Starting the dev server

From your agent directory, run:

selu dev .

This command:

1. Validates agent.yaml and agent.md.
2. Builds any local capability Docker images.
3. Starts the orchestrator with your agent loaded.
4. Opens a web chat interface at http://localhost:4840.

Tip

Add --watch to auto-reload on file changes: selu dev --watch .

Viewing logs

All orchestrator and capability logs are streamed to your terminal. You can also inspect them with:

selu logs                  # all logs
selu logs --capability weather   # filter by capability name
selu logs --level debug    # include debug output

Capability containers write to stdout/stderr, which the orchestrator captures and prefixes with the capability name.

Testing with the CLI

For quick, non-interactive testing, use selu chat:

selu chat . "What's the weather in Berlin?"

This sends a single message, prints the agent’s response (including any tool calls), and exits. Useful for scripting and CI.

Snapshot testing

Selu supports snapshot tests — pre-recorded conversations that verify your agent’s behaviour doesn’t regress.

Create a snapshot

selu test snapshot create --name "greeting" .
# Interactive chat session opens — type your test conversation
# Press Ctrl+D to save

Run snapshots

selu test snapshot run .

The CLI replays each recorded conversation, compares tool calls and response structure, and reports pass/fail.

Note

Snapshot tests check structural correctness (did the agent call the right tool with the right arguments?) rather than exact text matching, since LLM outputs are non-deterministic.

Debugging capabilities

If a capability isn’t behaving as expected:

1. Check health — selu status shows the state of each running capability container.
2. Manual gRPC call — Use grpcurl to invoke the capability directly:

   grpcurl -plaintext -d '{"tool_name": "weather_lookup", "parameters": "{\"location\": \"Berlin\"}"}' \
     localhost:50051 selu.capability.v1.CapabilityService/Invoke

3. Shell into the container — docker exec -it selu-cap-weather /bin/sh

Environment variables

Capabilities that need API keys or configuration can read from environment variables. Set them in a .env file in your agent directory:

.env

WEATHER_API_KEY=your-api-key-here

The dev server automatically loads .env and passes the variables to capability containers.

Next steps

Once your agent works locally, review the Marketplace Guidelines to prepare for publication.