keakon's recent timeline updates keakon's recent timeline updateskeakon's repos on GitHub
Go · 39 watchers
golog
Super fast, unstructured logging in Go.
Python · 35 watchers
Doodle
A blog system based on Python 2.7 and Redis.
Go · 12 watchers
chord
A lightweight, local-first terminal coding agent. Low resource usage, reliable long sessions, flexible model orchestration, Vim-like navigation, multi-agent collaboration, and a headless mode for remote control — designed to make AI coding feel calm and predictable.
JavaScript · 9 watchers
Copy-Unlock
Python · 6 watchers
claude-code-cost
A Python tool for analyzing Claude Code usage history, calculating token consumption and costs across projects and time periods.
Python · 3 watchers
etag-middleware
A middleware to handle ETag for Starletter or FastAPI.
Python · 3 watchers
fastapi-skeleton
A web application template of FastAPI with best practice.
Kotlin · 2 watchers
ClashMetaForAndroid
A rule-based tunnel for Android.
2 watchers
NaiveProxy
一个不占用内存的 Chrome 代理扩展。
Go · 1 watchers
chord-gateway
chord-gateway connects IM platforms such as WeChat and Feishu to local chord headless processes. It lets you control Chord from chat while keeping the agent process, workspace access, credentials, and state on your own machine or server.
Python · 1 watchers
delayed
A simple but robust task queue.
1 watchers
go-delayed
A simple but robust task queue.
Go · 1 watchers
go-utils
Python · 1 watchers
gunicorn
gunicorn 'Green Unicorn' is a WSGI HTTP Server for UNIX, fast clients and sleepy applications.
TypeScript · 1 watchers
TraceViewer
Real-time viewer for Claude Code / Codex agent traces
0 watchers
bubbles
TUI components for Bubble Tea 🫧
0 watchers
bubbletea
A powerful little TUI framework 🏗
Python · 0 watchers
huey
a little task queue for python
Go · 0 watchers
hysteria
Hysteria is a powerful, lightning fast and censorship resistant proxy.
0 watchers
mihomo
A simple Python Pydantic model for Honkai: Star Rail parsed data from the Mihomo API.
0 watchers
oh-my-opencode
The Best Agent Harness. Meet Sisyphus: The Batteries-Included Agent that codes like you.
0 watchers
opencode
The open source coding agent.
0 watchers
opencode-antigravity-auth
Enable Opencode to authenticate against Antigravity (Google's IDE) via OAuth so you can use Antigravity rate limits and access models like gemini-3-pro and claude-opus-4-5-thinking with your Google credentials.
TypeScript · 0 watchers
opencode-dynamic-context-pruning
Dynamic context pruning plugin for OpenCode - intelligently manages conversation context to optimize token usage
Python · 0 watchers
peft
🤗 PEFT: State-of-the-art Parameter-Efficient Fine-Tuning.
Python · 0 watchers
redis-shard
Redis sharding client library
0 watchers
seahub
The web end of seafile server.
Python · 0 watchers
selectors2
Backported, durable, and portable selectors designed to replace the standard library selectors module.
Go · 0 watchers
shard
0 watchers
sing
Do you hear the people sing?
Go · 0 watchers
sing-quic
quic-go API wrapper and QUIC based protocol implementations
Go · 0 watchers
sql-builder
Go · 0 watchers
sql-to-struct
Python · 0 watchers
starlette
The little ASGI framework that shines. 🌟
Python · 0 watchers
taskiq
Distributed task queue with full async support
Python · 0 watchers
tornado
Tornado is an open source version of the scalable, non-blocking web server and tools that power FriendFeed.
Jupyter Notebook · 0 watchers
train-qwen
Go · 0 watchers
ultraviolet
Mystical terminal user interface primitives 🌈
Python · 0 watchers
uvicorn
The lightning-fast ASGI server. 🦄
Python · 0 watchers
white
A fork of the Black formatter but prefers single quotes.
0 watchers
wxocr
This project wraps the WeChat OCR functionality from the excellent wechat-ocr project into a simple REST API service that can be easily deployed using Docker. It allows you to perform optical character recognition on images by leveraging WeChat's powerful OCR capabilities.
0 watchers
x
Charm experimental packages
 |
keakonV2EX member #2704, joined on 2010-11-01 02:41:26 +08:00 |
keakon
keakon.top
keakon
| MiniMax M2.1 感觉已经追上 Claude Sonnet 4.5 了 Claude • keakon • Dec 29, 2025 • Lastly replied by wayhome | 12 |
| Cursor 本周 GPT-5 系列模型免费 Cursor • keakon • Aug 8, 2025 • Lastly replied by sdjl | 7 |
| Gemini CLI 号称的每天 1000 万 tokens 和 1000 次请求 Google • keakon • Aug 9, 2025 • Lastly replied by realpg | 13 |
| 坚持到重置前的最后一天了 Cursor • keakon • Aug 5, 2025 |
| 用 Claude Code 做了个 Claude Code 的成本计算器 Claude • keakon • Jul 29, 2025 • Lastly replied by keakon | 13 |
| 没有买 Cursor 的用户还是用 Claude Code 吧 Claude • keakon • Aug 13, 2025 • Lastly replied by jasonStark2 | 9 |
| Cursor 澄清新定价方案了 Cursor • keakon • Jul 14, 2025 • Lastly replied by alexluo1 | 11 |
| AI 帮我重新上架了 Chrome 插件： Copy Unlock 分享创造 • keakon • Apr 18, 2025 • Lastly replied by keakon | 17 |
| 5 年小米/红米用户，准备换新手机，问下其他厂商有啥推荐的 Android • keakon • Dec 3, 2024 • Lastly replied by spark | 75 |
| 如何让 Rime 在双拼时，优先用 2 个字母确定一个汉字？ 中州韻 • keakon • Nov 22, 2024 • Lastly replied by isudfv | 3 |
keakon's recent replies
23h 27m ago Replied to a topic by coder121 › 程序员 › GPT sol 的上下文长度设置成多少比较合适，默认的 200k 太短了，随便几下就开始合并了， 1m 到后面又有点顾头不顾尾的 |
codex 有个功能是让模型自行记录笔记，并自主进行上下文切换。
我把这个功能移植到了自己的 agent ，感觉挺好用的，模型会在完成子任务或上下文较多时就瞬间清理上下文，可以无缝连续长程工作。我有个任务已经跑了 8 小时了，推进仍然很顺利。
我把这个功能移植到了自己的 agent ，感觉挺好用的，模型会在完成子任务或上下文较多时就瞬间清理上下文，可以无缝连续长程工作。我有个任务已经跑了 8 小时了，推进仍然很顺利。
Aug 31 Replied to a topic by IC0ZB › 程序员 › 记一次 GLM 开发严重事故，花了 8 亿 token，买了个教训 |
我是自己做 harness ，rm 和 git reset 等危险的操作需要手动确认，避免了很多问题
Aug 28 Replied to a topic by Diodeme › 程序员 › 只有 33 个 star 的 Agent 客户端被 OpenAI 送 1200 美元，我们是怎么做到的？ |
申请通过后，好像还是要国外信用卡才能开通吧？
Aug 27 Replied to a topic by TirionHo › 程序员 › Luna Max 有这么强吗？ |
比 deepseek-v4-flash 稍好但不多，评价高的原因是便宜
Aug 24 Replied to a topic by jangit › 程序员 › 分享一个移动端 vibe code 方案： dinotty |
@jangit 我试了下没收到通知，甚至配置里直接播放声音也听不到，还不知道原因
Aug 23 Replied to a topic by jangit › 程序员 › 分享一个移动端 vibe code 方案： dinotty |
通知是否能接入 OSC 9/777 ？
楼主算的价格和其他人差了一个数量级，帮你验算一下。
24.3 亿 是 2430M tokens ，全按缓存来算 $0.5/1M 是 $1215 ，和你的 $1250 是很接近的，多余的部分是没有缓存或是 output ，这比较符合实际。
1250/0.65 = 1923 ，这也符合大家测出来 $2000 左右的周额度。
中转站主要问题是缓存利用率不高且很容易造假，同样的 1 亿 token ，如果缓存命中率 99%，费用是%54.5 ，如果是 98%，费用是 59 ，能相差 10%。大部分中转站能到 92%就不错了。
24.3 亿 是 2430M tokens ，全按缓存来算 $0.5/1M 是 $1215 ，和你的 $1250 是很接近的，多余的部分是没有缓存或是 output ，这比较符合实际。
1250/0.65 = 1923 ，这也符合大家测出来 $2000 左右的周额度。
中转站主要问题是缓存利用率不高且很容易造假，同样的 1 亿 token ，如果缓存命中率 99%，费用是%54.5 ，如果是 98%，费用是 59 ，能相差 10%。大部分中转站能到 92%就不错了。
Aug 10 Replied to a topic by YuanJiwei › 程序员 › 做了一款终端优先的 AI 编程环境 Termio.sh, 比 Ghostty/ iTem2 / cmux 好用 |
ctrl+t 不是打开当前会话的路径，而是打开 home 路径，这个也不适应
Aug 10 Replied to a topic by YuanJiwei › 程序员 › 做了一款终端优先的 AI 编程环境 Termio.sh, 比 Ghostty/ iTem2 / cmux 好用 |
Aug 9 Replied to a topic by fltv › 推广 › [薅羊毛] mirasim（注册账号赠送额度：等于 Codex Pro 5× + Claude Code Max 5×） |
MIRA-H1J3-NBS6-SM36
感觉有点卡
感觉有点卡
Select Language