Write important state once, then recall it before risky work.

Commit only what should become durable memory.

from bilinc import CloudClient

client = CloudClient()
client.commit(
    "project.status",
    {"phase": "docs refresh", "verified": True},
    memory_type="episodic",
    importance=0.8,
)

Recall before continuing a workflow.

results = client.recall(
    "what did we verify in the docs refresh?",
    profile="balanced",
    limit=5,
)

Treat failures as non-durable.

If the Cloud API does not return success, the agent should behave as if the write did not happen. This keeps durable memory honest.