Auto-artifact log
Every Edit/Write becomes an immutable event in logs/devlog.sqlite. Zero manual log_event calls. Two-hook design ships in 5 minutes.
Stop re-explaining your project to the AI.
Codetrail turns Claude Code into a persistent project partner: every file edit logged automatically, every use case & test case tracked, stale-context warnings on resume. No more confident-wrong narratives.
MIT
open source license
54
events in demo devlog
local-first
no telemetry
0
bytes uploaded for viewer
codetrail.dipgle.com/viewer.html?demo=1
Events
Use cases
Test cases
Test runs
| ID | TS | Kind | Ref | Content |
|---|---|---|---|---|
| 54 | 2026-06-02 14:21 | artifact | file:crates/server/src/main.rs | Write: bind /healthz JSON envelope |
| 53 | 2026-06-02 14:18 | decision | — | Use rmcp 0.16 (bundled rusqlite, single binary) |
| 52 | 2026-06-02 14:15 | artifact | file:Cargo.toml | Edit: add chrono + schemars deps |
| 51 | 2026-06-02 13:58 | test_run | TC-007 | RED — getContextBrief returns null on empty DB |
| 50 | 2026-06-02 13:52 | message | actor:user | "split scan_health into queryable signals" |
UC / TC tracking, TDD-aligned
Use cases and test cases live in the same SQLite next to every event. RED → GREEN runs are recorded, not implied. Session resume reads from data, not vibes.
Stale-resume warning
SessionStart hook compares devlog last-event with file mtime. If code moved without logging, the warning lands directly in the prompt — impossible to miss.
Multi-project memory
5, 20, 50 projects in one workspace? Codetrail walks up from any file to find the right devlog. One brain across all your codebases.
Browser-native viewer
Drag any devlog.sqlite into the viewer — events, UCs, TCs, test runs render instantly. Pure WASM, your data never leaves the browser.
Local-first, no cloud
SQLite stays on your disk, alongside the code. There is no cloud component to opt in to. Privacy by default, not by preference toggle.
Why Codetrail
Honest table. We track what no other AI-coding tool tracks structurally.
| Codetrail | Claude Code | Aider | Cursor | |
|---|---|---|---|---|
| Structured project devlog (events / decisions) | ● | — | git only | — |
| UC / TC + test_runs tracked separately | ● | — | — | — |
| Multi-project memory across N codebases | ● | partial (chat memory) | per-repo | per IDE session |
| Auto-log every artifact without code calls | ● | discipline-based | inherent | — |
| Browser-native viewer for any team member | ● | — | terminal log | — |
| Stale-resume warning at session start | ● | — | — | — |
| Local-first; no cloud component | ● | hybrid | ● | cloud-only |
| Setup time | ~30s (np my-app) | ~5min | ~1min | ~10min IDE config |
Spotted something off? Tell us — we update this table when we're wrong.
How it works
-
Install
One bash helper:
np my-appscaffolds CLAUDE.md, docs/, memory/, logs/devlog.sqlite, MCP server. Oradoptan existing project in place. -
Work normally
Open Claude Code. Auto-hooks log every artifact. The MCP server tracks UCs/TCs/test runs as you build. Nothing extra to remember.
-
Resume any time
Days, weeks, multiple agents later — Codetrail surfaces the right context. The narrative comes from data, not stale notes.
Install in 30 seconds
Two helpers: np for new projects, adopt for existing ones. Works on macOS, Linux, Windows (native or WSL2). Requires Node 20+ and Claude Code.
1
Add the helpers to your shell
curl -fsSL https://codetrail.dipgle.com/install.txt | bash
source ~/.zshrc2
Start a new project
np my-app # scaffolds + opens Claude Code
# or adopt an existing project:
cd /path/to/repo
adopt3
That's it
PostToolUse and SessionStart hooks register globally. Every Edit/Write auto-logs. Open Claude Code in any project — Codetrail runs in the background.
Honest answers to the questions you'd actually ask
How is this different from Claude Code's built-in auto-memory?
Auto-memory is chat-style — it summarizes conversations into prose. Codetrail is structural: every file edit becomes an artifact event with path, timestamp, and tool. Every decision is a typed decision row. Every test run records pass/fail with notes. You can query, filter, and audit. The two are complementary; we plug into hooks while auto-memory runs at the model layer.
Does my code leave my machine?
No. Hooks write to logs/devlog.sqlite on your disk. The web viewer parses SQLite in your browser via WASM — zero upload. Codetrail has no cloud component; everything runs locally. See the privacy policy.
What if Claude Code changes its hook API?
The hooks live in ~/.claude/settings.json and shell out to plain bash scripts that only depend on sqlite3 and Python 3. If the schema or matcher format changes, we ship an update — the migration is a one-line settings edit. The devlog format is independent and we'll always provide a converter.
Is the code open source?
Yes — everything is MIT. Hooks, landing, project template, and both MCP server implementations (Node + Rust). See github.com/dipgle/codetrail.
I already use Cursor / Aider — can I still use Codetrail?
Codetrail's hooks are Claude Code-specific (they hook into Claude Code's PostToolUse / SessionStart events). The browser viewer is editor-agnostic — drop any compatible SQLite there. Aider and Cursor integrations aren't on the roadmap but the schema is documented; community PRs welcome.
Open source on GitHub
Landing, hooks, project template, SDK — MIT-licensed. Drop a star if Codetrail saves you time; file an issue if it doesn't.
Stay in the loop
Codetrail is free and fully MIT-licensed today — clone the repo or run install.txt. Drop your email if you'd like a short note when a major feature ships. No spam, opt-out anytime.