Manage Environments

Testing before going live is crucial, and the same applies to your AI.

How it works

Every agent is seeded with two environments: development and production. They are created with the agent, and every conversation is recorded against one of them. An agent given further environments server-side keeps working unchanged — the SDK learns the full list at startup.

• Development: for testing and iteration.
• Production: for live, real-world interactions.

AgentSight tags each conversation with its environment, so test traffic and real traffic stay separate in your dashboard.

Rules & Behavior

Environment	Purpose	Dashboard Access	Analytics Availability
Development	Used for building and testing your AI.	View transcripts, messages, actions, and attachments.	❌ No analytics or reports (test data is excluded from insights).
Production	Used for deployed, real-world conversations.	Full access to dashboards, analytics, and reports.	✅ Analytics and reports available.

To view your development data inside the dashboard, simply toggle Development Mode. Analytics and reports are computed from production traffic only, so Development Mode shows you which metrics are receiving data instead of charts — enough to confirm your integration is working, without test runs skewing the numbers your dashboards report.

Note: Conversations are recorded against production unless you select an environment.

How to set the environment

Set it once for the whole deployment. Either pass it to init():

agentsight.init(environment="development")

or set it in the environment, which is usually what you want — the same code then records against development on a developer's machine and production once deployed:

AGENTSIGHT_ENVIRONMENT=development   # or: production

dev and prod are accepted as shorthand for either form.

Overriding it for one conversation

The deployment-wide setting is the default, not a ceiling. A single conversation can name its own environment, which is what you want for a seeded demo or an evaluation run inside a production process:

with agentsight.conversation("eval-run-41", environment="development"):
    ...

Everything recorded inside that conversation inherits it.

Checking what an agent has

environments() on the API client is the authoritative list:

from agentsight.api import AgentSight

ags = AgentSight()
print(ags.environments())   # ['production', 'development']

When the value is wrong

An unrecognised environment is reported as an error at startup, while you are still watching the logs — not discovered later as a batch that quietly went nowhere.

If one reaches a conversation anyway, that conversation is recorded without an environment rather than being dropped, and the SDK warns once naming the value it could not place. A typo costs you the environment tag; it never costs you the data.

For more configuration options, see the Configuration page.