Set Context

set_context() attaches custom key/value dimensions — customer ID, feature, plan, team, whatever you already tag internally — to the currently active trace. Two dashboard surfaces read these values back: Cost’s Segments tab and custom-metric SLOs on the Reliability screen.

Signature


morse_ai.set_context(**dims: str) -> None

Call it inside an active trace — typically once per request, e.g. in Django middleware:


morse_ai.set_context(customer_id=request.customer_id, feature="doc-qa")

Values are coerced to str() at the SDK boundary; non-string values you pass in are converted automatically.

Outside an active trace, set_context() is a silent no-op — it does nothing until called from inside morse_ai.run(...) or @morse_ai.trace_agent.

You can also pass the same dimensions directly when opening a trace, via run()’s user_metadata kwarg — set_context() is the mid-trace equivalent for values you don’t know yet at run() time:


with morse_ai.run(
    "chat-agent",
    thread_id=session_id,
    user_id=current_user.id,
    user_metadata={"customer_id": tenant_id, "feature": "doc-qa"},
) as r:
    ...

`thread_id` without any code change

When an adapter opens the trace for you, it fills thread_id in from whatever conversation identity the framework already carries — an openai-agents session= / conversation_id= or a thread/thread_id on the run context, a LangGraph configurable.thread_id, the Claude Agent SDK’s session id. A thread_id you pass to run() always wins; the adapter only fills a gap.

Both SDKs do this. In TypeScript the field is threadId on run() / runTraced(), and the adapters read the same places (session / conversationId / the run context for openai-agents, configurable.thread_id for LangChain and LangGraph, the message session_id for the Claude Agent SDK).

What reads these dimensions

Segments (Cost)

The Cost screen’s Segments tab breaks down cost, coverage, and trace count by any key you’ve set. It stays empty until your code calls set_context() at least once — the empty state prints the exact set_context() line to add, rather than hiding the capability.

Custom-metric SLOs (Reliability)

An SLO can target a custom metric instead of a built-in one (success rate, cost, latency). Defining one asks for a Key field labeled “matches your set_context() key” — the SLO evaluates whatever value you attach under that key on each trace.

Overview — the SDK’s tracing primitives (run(), span(), trace_agent).
Installation — minimal init() call and API key setup.
Cost — where Segments render the dimensions you set.
Reliability — where custom-metric SLOs consume the same key.