Agently DevTools Examples

These examples show the core DevTools integration paths from the main Agently repo.

Before running them:

pip install -U agently agently-devtools

Observation and evaluation examples also need the local DevTools listener:

agently-devtools start

For the interactive wrapper examples, starting the listener is also recommended when you want to inspect Agent or TriggerFlow runs in the DevTools console while using the browser UI.

Examples in this folder:

ObservationBridge (Passive monitoring):

• 01_observation_bridge_local.py Uses Agently.create_observation_bridge(...).watch(Agently) to upload one simple TriggerFlow run.
• 02_observation_bridge_selective_watch.py Uses Agently.create_observation_bridge(watched_flow, ...) so only the selected flow is uploaded.
• 03_scenario_evaluations.py Uses EvaluationBridge and EvaluationRunner to run a small repeatable suite.
• 07_agently_observe_lazy_bridge.py Uses Agently.observe(flow, ...) to lazily load DevTools and watch one TriggerFlow without importing agently_devtools in app code.
• 08_model_request_telemetry_local.py Registers a local Event Center hook and shows payload.model_request_telemetry on model request RuntimeEvents; this is an infrastructure probe and does not require a real model provider.

InteractiveWrapper (Active interaction):

• 04_interactive_wrapper_basic.py Uses InteractiveWrapper() with a generator callable to stream text chunks into the interactive chat UI, and also shows how to register the example with ObservationBridge.
• 05_interactive_wrapper_agent.py Uses InteractiveWrapper() with an Agently Agent so token output can stream into the UI when the configured model supports it, while Agent runs are also published to DevTools via ObservationBridge.
• 06_interactive_wrapper_trigger_flow.py Uses InteractiveWrapper() with a TriggerFlow that emits stage updates before returning the final structured result, while TriggerFlow runs are also published to DevTools via ObservationBridge.

Default local listeners:

ObservationBridge & EvaluationBridge (Devtools workbench):

• Console: http://127.0.0.1:15596/
• Ingest: http://127.0.0.1:15596/observation/ingest

InteractiveWrapper (Interactive demo UI):

• Default port: 15365 (similar to AGENT key positions)
• General usage: python <example_file>
• Access the interactive UI in your browser: http://127.0.0.1:15365/
• The built-in page uses /api/stream first and falls back to /api/chat only when the wrapped provider does not support streaming.
• 04_interactive_wrapper_basic.py keeps the provider as a plain callable, so it focuses on wrapper streaming; 05 and 06 are the examples that also produce Agently runtime observation data in DevTools.

All examples avoid external model dependencies so they can be used as integration smoke tests.