AI Decision System
Replayable, offline-verifiable records for AI-assisted go/no-go decisions: experiment promotions, typed model answers and human-approved actions.
翻譯準備中
尚未偵測到更新。GitHub 發布版本後會顯示在這裡。
個人 GitHub 帳號下公開儲存庫的擁有者可以登入並驗證所有權,自行管理介紹,無需管理員核准。
登入並產生驗證檔案,在 30 分鐘內將其提交至此儲存庫的預設分支。其他分支或分叉儲存庫中的檔案無效。然後在下方檢查檔案,即可無需管理員核准直接管理專案。請保留檔案以維持管理權限。
如需登入,將前往登入頁面。登入後請回到此處,再次點擊所選的驗證按鈕。
在專案資料夾中執行 /nomorevibe verify,即可驗證所有權並自行管理介紹。
一分鐘完成驗證Trainthon - 24시간 기차 해커톤 스타터 킷
翻譯準備中
取自儲存庫 README
正在載入完整 README。 以下為已儲存的摘錄。
Remember what your agents already found.
Local-first memory over Claude, GPT/Codex, and Cursor runs on this machine.
Archive the original logs, collapse them into what each step obtained, and keep the DAG as the receipt.
Agents forget each other. Claude’s session does not see Codex. Cursor does not see last week’s Claude run. DeepTracer keeps those logs on this computer, compresses tool calls into meaning, and is the store you query next.
It is not a JSON paste box. You register desktop agents. A local Next.js server reads session files here:檢視完整 README ↗Agents forget each other. Claude’s session does not see Codex. Cursor does not see last week’s Claude run. DeepTracer keeps those logs on this computer, compresses tool calls into meaning, and is the store you query next.
It is not a JSON paste box. You register desktop agents. A local Next.js server reads session files here:
Register agents → archive sessions → meaning graph → recall → receipt DAG
A hosted Vercel/Netlify site cannot open ~/.claude, ~/.codex, or ~/.cursor. That is the product, not a limitation to work around.
flowchart TB
O[Orchestrator]
O --> R[Research]
O --> B[Browser]
R --> S[Vendor A pricing]
B --> T[Pricing table]
Nodes are what a step obtained — a source, a file, a page — not Search ×4. The graph is the compressed memory. Open a node when you need the original.
| Agent | Desktop app | Logs |
|---|---|---|
| Claude | Claude | ~/.claude/projects, ~/.claude/transcripts |
| GPT | ChatGPT / Codex | ~/.codex/sessions |
| Cursor | Cursor | ~/.cursor/projects/*/agent-transcripts |
Only installed desktop apps are listed. Credentials, auth files, and SQLite chat DBs are ignored.
git clone https://github.com/acradin/trainthon.git
cd trainthon/deeptracer
npm install
npm run dev
Open http://localhost:3000. Register the agents on this machine. Runs appear without uploading JSON.
Optional deeptracer/.env.local (do not commit):
OPENAI_API_KEY=
OPENAI_MODEL=gpt-5.6-luna
Review intent and meaning-graph compression use OpenAI when a key is present. Without it, traces still scan and open. Default storage is ~/.deeptracer.
| Path | Role |
|---|---|
/ | Product landing (Recall-first). Registered visits go to /recall |
/onboard | Register desktop agents |
/recall | Ask what this machine already found |
/dashboard | Runs, grouped by project |
/trace/[id] | Meaning graph (receipt), Inspector, Timeline |
/agents | Status + Scan now |
/analyze | Example traces and intent review |
/import, /setup | Manual JSON / OTLP fallbacks |
npm run dev.App code lives in deeptracer/.
Built for Cursor Korea × Yonsei University Startup Support Group Trainthon.