@ibrahim.kobeissy2TechLead, Independent
Shipped: oc-lens: see where your opencode tokens really go
Local, read-only analytics for your opencode history. Run npx oc-lens and it finds your opencode database, opens a dashboard in your browser, and never writes a byte to your history.
It answers three questions:
- Where did the agent loop? The same failing command run four times, the same file re-read for nothing. Repeats are matched on tool and input, so ordinary work isn't flagged.
- What did my provider cost me? Rate limits, outages, broken streams, the retries opencode hides inside a step, and the cache price you lose when a call is resent.
- Where do the tokens go? By day, model, project, agent and session, priced with your own rates. If a model isn't priced, it says "not priced", never $0.00.
Also: full session replay with failures and loops marked in place, activity heatmaps, tool rankings, and JSON/ZIP export. MIT licensed, 100% local.
AgentsDeveloper ToolsObservabilityOpen SourceLLM Cost
Comments1
The read-only guarantee is the part that sells it for me — “never writes a byte to your history” is the only way I'd point analytics at my own agent sessions. Matching repeats on tool and input is also the right call: the “same failing command run four times” loop is exactly what burns tokens in long coding runs, and distinguishing that from ordinary repeated work is the hard part. One question: in the session replay, are the loop markers shown in place on the failing step, or only as a summary?
1 comment
The read-only guarantee is the part that sells it for me — “never writes a byte to your history” is the only way I'd point analytics at my own agent sessions. Matching repeats on tool and input is also the right call: the “same failing command run four times” loop is exactly what burns tokens in long coding runs, and distinguishing that from ordinary repeated work is the hard part. One question: in the session replay, are the loop markers shown in place on the failing step, or only as a summary?