Turnlog vs. the alternatives, fairly

The short version

Everything below reads the same local files (the JSONL in ~/.claude/projects/) and none of it uploads your sessions anywhere. The differences are in shape: usage reporting vs. browsing vs. indexing, CLI vs. desktop app vs. localhost server. Pick by the question you actually ask most.

ccusage — the usage-stats standard

A fast CLI that turns session logs into token and cost reports — daily, monthly, per-model, billing-window views — across a long list of agents (Claude Code, Codex, Gemini CLI, and many more), with JSON export and even an MCP server for usage questions. If you want usage reporting, use ccusage; Turnlog doesn’t replace it. The one cost view Turnlog adds is different in kind: cost filtered by a content search (“what did the websocket work cost?”), which needs a full-text index rather than a report generator.

Claude Code History Viewer — the desktop app

A polished desktop application for browsing history, with search, filters, and an analytics dashboard — and, notably, the widest agent coverage in the field: as of mid-2026 it reads Claude Code, Gemini CLI, Codex, Cline, Cursor, Aider, and OpenCode. If “every agent I use, one desktop app” is your requirement, it’s the strongest option today — Turnlog reads two agents (Claude Code, Codex), with more on the roadmap. Turnlog’s trade runs the other way: no app install (npx and a browser), and depth over breadth — the turn spine with lenses and bookmarks, cross-session file history, spend-by-search, redacted exports, and agent-side memory over MCP, which no viewer offers.

claude-code-trace — the flexible viewer

A session log viewer that meets you wherever you like your interfaces — GUI app, web, or TUI — with live tailing of running sessions. A good fit for “pretty-print this transcript, watch it move.” The difference is the one Turnlog’s whole pitch hangs on: a viewer renders files; an index imposes structure across all of them — cross-history search with operators, one timeline per repo across agents, resumed sessions stitched into one conversation, and cost attribution. (Turnlog also live-tails — the UI updates about a second after your agent acts.)

claude-history — the fuzzy finder

A deliberately small CLI: fuzzy-search your conversations — tool results included — jump to a session by UUID. If you want one lightweight thing in the terminal and no server at all, it’s a lovely tool. Turnlog’s CLI search (turnlog search "…") covers the same ground with the operator language, at the price of an index build on first run.

When Turnlog is the wrong choice

• You want usage dashboards and nothing else → ccusage.
• You want one desktop app across five-plus agents today → Claude Code History Viewer.
• You want a TUI transcript viewer or the smallest possible tool → claude-code-trace or claude-history.
• You want team-level or cloud analytics → none of these; and Turnlog will never build it.

What only Turnlog does

The combination Turnlog exists for: an FTS5 content index over your whole history with composable operators (tool: is:error agent: path: tag:…), turn-spine replay that collapses a 5,000-event session to its prompts, one project timeline where Codex and Claude Code work on a repo land together, spend by search query, cross-session file history, exports with a real redaction pass — and the index served back to the agent itself as read-only MCP memory. Hardened localhost server (loopback-only, per-launch token), two dependencies, MIT, no catch. You can try the whole thing in a browser demo before installing anything.

Related: searching your history · cost tracking, honestly · agent memory over MCP