SVSANNVIT
SDKs

Python SDK

Install and use the ledger's Python SDK — manual capture and one-line Anthropic/OpenAI auto-instrumentation, stdlib only.

The Python SDK (sdk/python, package ledger_sdk) has full parity with the TypeScript reference — manual capture plus Anthropic/OpenAI auto-instrumentation — and depends on nothing outside the standard library.

Install

Not yet published to PyPI. Vendor sdk/python/ledger_sdk into your project for now, or install it as a local/git dependency:

pip install <path-to-repo>/sdk/python

Quick start — manual capture

from datetime import datetime, timezone
from ledger_sdk import Ledger

ledger = Ledger(
    api_url="http://127.0.0.1:4010",
    ingest_key="...",
    agent_id="urn:agent:research-assistant",
    on_error=lambda msg: print("[ledger]", msg),
)

t0 = datetime.now(timezone.utc)
# ... do the work ...
t1 = datetime.now(timezone.utc)

ledger.record(
    action_type="llm_call",
    provider="anthropic",
    model="claude-fable-5",
    started_at=t0,
    ended_at=t1,
    status="success",
    usage={"tokens_in": 420, "tokens_out": 180},
    payloads=[
        {"kind": "prompt", "content": "Summarize the policy for the client memo."},
        {"kind": "response", "content": "The policy provides that…"},
    ],
    trigger={"type": "human", "subject": "person@example.com"},
)

ledger.flush()
ledger.close()

close() stops the background flush thread after a final flush — call it at shutdown so buffered events aren't lost.

action() — time and record in one call

result = ledger.action(
    {"action_type": "tool_call", "config": {"tool": "save_memo"}},
    lambda: save_memo(path),
)

Records success or failure automatically and re-raises the original exception untouched on failure.

Quick start — auto-instrumentation

from ledger_sdk import Ledger
from ledger_sdk.instrument import wrap_anthropic
import anthropic

ledger = Ledger(api_url="http://127.0.0.1:4010", ingest_key="...", agent_id="urn:agent:research-assistant")
ledger.set_trigger("human", "person@example.com")

client = wrap_anthropic(ledger, anthropic.Anthropic())

# unmodified application code from here down — no ledger references needed
res = client.messages.create(
    model="claude-fable-5",
    max_tokens=1024,
    messages=[{"role": "user", "content": "Draft the client memo."}],
)

wrap_openai(ledger, client) works the same way for OpenAI's client.

Decision lineage

span = ledger.record(action_type="search", started_at=t0, ended_at=t1, status="success")

with ledger.within(span["span_id"]):
    # anything recorded in here defaults its parent_ref to the search span
    client = wrap_anthropic(ledger, anthropic.Anthropic())
    client.messages.create(...)

Pass parent_ref explicitly on a record() call to override the ambient parent.

Uploading blobs

blob_id = ledger.upload_blob(file_bytes, "application/pdf")

ledger.record(
    action_type="data_read",
    payloads=[{"kind": "artifact", "blob_ref": blob_id}],
    started_at=t0, ended_at=t1, status="success",
)

OpenTelemetry

provider.add_span_processor(ledger.otel_span_processor())
# or: BatchSpanProcessor(ledger.otel_exporter())

Every gen_ai.* span becomes a ledger event; other spans pass through untouched.