Jingbiao's Blog

Tokdash: one dashboard for every AI agent

Tokens, cost, cache-hit rate, session drill-down, and quota resets — local-first, MIT, pipx install

If you drive more than one AI coding agent from the terminal — Codex, Claude Code, Antigravity, OpenCode, Gemini CLI, Pi, Kimi, the rest — you already know the problem: each one writes its own sess...

Posted by Jingbiao on July 3, 2026, 12 minutes to read

3-minute Pitch: ATM-Bench — According To Me, A Benchmark for Long-Term Personalized Memory QA

The First Benchmark Built From Four Years of One Person's Real Life

Project page here Paper on arXiv Code available here Dataset on Hugging Face Ask your AI assistant a question that should be trivial — “What was the gift I bought for my mom during our J...

Posted by Jingbiao on June 22, 2026, 12 minutes to read

Tokdash v1.0: A Local Token & Cost Dashboard That Installs, Runs, and Uninstalls Itself

The v1.0+ story: pipx → setup → doctor → update → uninstall, with overview, sessions, 3D heatmap, ten themes, and a live statusline hook — all on localhost

For the last six months every working session on my laptop has been a negotiation with at least one AI coding agent — Claude Code for refactors, Codex for quick prototypes, OpenCode when I want an ...

Posted by Jingbiao on June 22, 2026, 22 minutes to read

Your Agent's Work Belongs on a Web Page

work-canvas: a Claude Code skill, distilled from 2,000 of my own agent sessions

A long agent run just finished. Four hours, three subsystems, a dozen course-corrections. Now I’m scrolling up through thousands of lines of terminal output trying to answer three simple questions:...

Posted by Jingbiao on June 20, 2026, 6 minutes to read

Giving an Agent a Search Engine It Actually Owns

Why I self-host SearXNG + Firecrawl for my coding agents — and the tutorial is one prompt long

Wire a coding agent up to a third-party model — Kimi K2 via OpenRouter, DeepSeek through your own Azure endpoint, Qwen on a vLLM box, anything behind a proxy or aggregator — and a quiet thing happe...

Posted by Jingbiao on June 20, 2026, 14 minutes to read

The ATM-Bench Leaderboard is Live

One Table for Every Personal-Memory System

When we released ATM-Bench earlier this year, the most common piece of feedback we got was not about the dataset or the metrics. It was a question: “OK, but where can I see how my system stacks...

Posted by Jingbiao on May 22, 2026, 8 minutes to read

The Bitter Lesson of Personal Memory: Why OpenClaw Still Can't Remember You

Introducing ATM-Bench: The First Benchmark for Long-Term Personalized Memory QA

Paper: arxiv.org/abs/2603.01990 Code: github.com/JingbiaoMei/ATM-Bench Dataset: huggingface.co/datasets/Jingbiao/ATM-Bench Project Page: atmbench.github.io Ask your AI assistant or OpenC...

Posted by Jingbiao on March 16, 2026, 9 minutes to read

3-minute Pitch: ExPO-HM — Learning to Explain-then-Detect for Hateful Meme Detection

Published in ICLR 2026

Project page here Paper on arXiv / OpenReview Code available here Hateful meme detection has largely been framed as binary classification: a model outputs hateful or benign and, at best, a...

Posted by Jingbiao on January 22, 2026, 3 minutes to read

3-minute Pitch: Robust Adaptation of Large Multimodal Models for Retrieval Augmented Hateful Meme Detection

Published in EMNLP 2025 Main Conference (Oral)

Project page here Paper on arXiv / ACL Anthology Code available here In our earlier work RGCL (ACL 2024), we showed that retrieval guidance over a CLIP-style encoder could sharpen hateful ...

Posted by Jingbiao on September 13, 2025, 4 minutes to read

3-minute Pitch: Retrieval Guided Contrastive Learning for Hateful Memes Detection

Published in ACL2024 Main Conference

Project Page here Code available here Hateful memes are captioned images promoting hostility towards specific social groups. Most hateful memes detection systems are logistic classifiers bui...

Posted by Jingbiao on November 15, 2023, 5 minutes to read