The description names exactly what the workspace records and then names the property that matters more than any of it: the log only ever grows. What an agent did is a fact you can go back to, rather than a state you have to trust.
Most systems treat an approval as a gate and then forget it happened. Filing it in the same log as the tool call is what makes a run reconstructable afterwards.
How a run ended is recorded next to what it did. That is the line most agent traces are missing when you go looking for why something stopped.
No schema, no format, no retention window, no throughput. The trending description publishes five event names and one property, so that is all the ledger carries.
Concept mock, not affiliated with the project. Every entry in the ledger is quoted from the repository's own GitHub Trending description on 23 August 2026: "Apache Maka (Incubating) is a local-first AI agent workspace. Model messages, tool calls, tool results, permission decisions, and termination events are recorded as an append-only log." Sequence numbers are positional only; no sample run, timestamp, or payload is represented here.