AGANAL reads the session logs your tools already leave on disk and turns them into metrics — tokens, tool calls, context pressure, retries — for every agent, on one screen.
Live on Product Hunt
AGANAL — agentic sessions analysis
Come say hi & leave feedback
A narrated tour — dashboard · events · raw JSONL · CLI · agent hand-off
The whole run as a dashboard, a filterable event stream, and the raw record — plus a one-click prompt (and a CLI) to hand it to an agent.
Scrub the run as a stream — messages, reasoning, tool calls and their results (paired by id), token usage per turn, lifecycle. Filter by type, search the text, jump to errors, or drag the event-range slider down to a single slice.
When you want the ground truth, drop to Raw JSONL: every line as a typed row — user, assistant, attachment, tool — and the exact record behind whatever you select, syntax-highlighted.
A second Claude install, a synced cloud Codex instance, an archive — point AGANAL at the folder, choose the format (Claude Code, Codex, Gemini CLI, Qwen Code, Cursor, opencode or Antigravity), and it reads straight into the same model.
Turn any session into a ready-to-paste prompt — pick a kind (Analyze, Find errors, Cost review, Critique, Continue) and it fills in the session plus the aganal commands. Or drive the CLI yourself — aganal analytics <path>, events, search — JSON an agent can load and reason over.
$ git clone https://github.com/the-ihor/aganal $ cd aganal && swift run ▸ discovering sessions… 7 sources · 214 sessions ▸ building normalized model… ▸ ready. opening AGANAL ✦
No installer, no account. swift run builds the native app.
Your agents are auto-discovered — or drop in any directory of logs.
Charts, tool calls, context and every event — from files already on disk.
AGANAL reads the logs your tools already wrote to disk and computes every metric locally. Nothing is uploaded, there’s no account, and there’s no telemetry — it works fully offline.
macOS 14+ · Apple Silicon & Intel · notarized · free forever
AGANAL