1.4 Short Circuit2026-10-21
Resume Matcher 1.4: Short Circuit
Short Circuit makes the resume itself better, not only better matched. Every tailoring run now critiques its own writing and rewrites the weak lines without adding a single fact, you can push a missing job keyword in when you confirm you have it, and you can watch each stage of the work as it happens. It also brings agent mode, an MCP server, new templates and multi-page resumes, and a round of security hardening.
Named after Short Circuit from Daft Punk's Discovery, this release continues the v1 series after 1.3 Crescendolls.
What's new
Writing that holds up
- Critique and rewrite in every tailoring run. Fixed rules and an AI critic check each bullet and the summary. Lines with an em dash, a weak opener such as "Responsible for", or no stated outcome get one rewrite. A fact check drops any rewrite that adds or loses a number, name, tool or skill, so your work stays on the page and nothing is invented. Light nudge only critiques.
- Writing score. A 0 to 100 score from the fixed rules (the same with every AI provider), with before and after, the lines to fix and the reviewer's notes.
- Improve writing in the Builder. A new Writing tab scores any resume, master included, and rewrites its weak lines. You review every change before it goes into the editor.
- Heavy rewrite. A fourth tailoring mode for bigger edits: full Challenge, Action, Result rewrites, a new summary and reordered bullets, with a clear hallucination warning.
- Evidence-based cover letters. One page, usually 200 to 300 words: a hook tied to the job, two or three matching examples, a company-fit paragraph grounded in the job description and a specific close. Contributed by @AlexanderMorris2k.
Keywords on your terms
- Add missing keywords. In the review dialog, missing job keywords become chips. Keywords your master shows are applied directly; a keyword it does not show needs your "I have this". Each one goes into your skills plus at most two lines where it fits, and nothing else is added around it.
- See where they went. The dialog lists where every keyword landed, marks it in the changed lines and shows the ATS score before and after. JD Match highlights the keywords you added in blue, and interview prep asks about them first.
- One match number. JD Match now shows the same keyword match as the ATS score, and keeps C++, .NET, Node.js and CI/CD whole.
- A sharper ATS score. It now scores the job title and date consistency, and the builder keeps the last calculated score with a Recalculate button. Contributed by @himalyansailor.
- ATS parse check. A new ATS CHECK tab shows whether a plain text extractor can read your exported PDF in the right order.
Watch it work
- Live progress. Tailoring shows each real stage as it runs, with a note when a slow model needs a minute.
- Tailoring queue. Queue several job descriptions and tailor them one after another, with cooldowns that respect rate limits. Contributed by @coolb0y.
Use it your way
- Agent mode. Open Claude Code, Codex, Cursor, Copilot or Antigravity in the repo and ask it to tailor your resume: no API key, and the web app updates live as the agent saves.
- MCP server. Its tools let any MCP client drive Resume Matcher against the same database as the app. Contributed by @rrvrs.
- Claude Code and ChatGPT plans as providers. Run on a Claude Code install with no API key (contributed by @GongYuanCaiJi), pick its model, or sign in with a ChatGPT plan, with a clear note on subscription risk.
- Optional PostgreSQL for larger or shared setups.
More control over the resume
- Several master resumes. Keep up to five, one per career track; tailoring picks the strongest bullets with code, not the AI.
- Page limit and bullet cap settings for tailoring. Contributed by @himalyansailor.
- New templates: Deedy, Modern Deedy and Vivid Single Column, plus multi-page resumes with predictable page breaks and an academic CV mode.
- The PDF matches the preview, and each resume keeps its own formatting.
- Reorder everything: projects, custom section items (contributed by @Bdysj) and the bullets within an entry.
Security
- Saved API keys can no longer be sent to arbitrary servers (#1045, reported and fixed by @MahediBuilds). An unauthenticated caller could point a stored key at their own server; a saved custom Base URL is now bound to the key that confirmed it.
- Local by default (#1048). The app has no login, so it now listens on loopback only and rejects unexpected Host headers.
- LiteLLM. Resume Matcher pins litellm==1.104.2 and was not affected by the compromised LiteLLM 1.107.0 release of 2026-10-21. Keep that pin in mind before upgrading LiteLLM.
- Dependency updates: cryptography 50.0.0, Next.js 16.3.3, DOMPurify 3.4.13, undici 7.30.0, sharp 0.35.4, js-yaml 4.3.2 and Vitest.
Upgrading from 1.3
Back up the complete data directory first, as always: the database, config.json, uploaded files and the hidden .secret_key together.
- The app now listens on 127.0.0.1 only. If you open it from another machine or a reverse proxy, set BINDADDRESS (Docker Compose) or HOST (source installs) and add the host name to ALLOWEDHOSTS, for example BINDADDRESS=0.0.0.0 ALLOWEDHOSTS=192.168.1.20 docker compose up -d. See SETUP.md for details.
- Database changes are additive and run on startup: new columns for ATS scores, per-resume formatting and forced keywords.
- Source installs still need Node.js 22+ and Python 3.13+; reinstall dependencies after pulling.
Docker images
Release tag: v1.4.0. Versioned images for Linux amd64 and arm64 publish to GHCR and Docker Hub:
``bash
docker pull ghcr.io/srbhr/resume-matcher:1.4.0
or
docker pull srbhr/resume-matcher:1.4.0
`
The image tags are 1.4.0 and 1.4`.
Verification
The release passed 1,944 backend tests and 1,596 frontend tests with type checking. The writing loop, live progress and keyword forcing were also run end to end against a real model through the app's proxy.
Thank you
Thank you to everyone who contributed code, reported bugs on Discord and GitHub, tested fixes and reviewed this release.
Contributors to this release: @576657922, @AlexanderMorris2k, @Bdysj, @coolb0y, @Frankli9986, @gingeekrishna, @GongYuanCaiJi, @himalyansailor, @HT-Moh, @kobihikri, @MahediBuilds, @rrvrs, @srbhr.
Automation contributions are credited to @dependabot.
<details>
<summary>All changes and first-time contributors</summary>
What's Changed
* build(deps): bump @vitest/mocker and vitest in /apps/frontend by @dependabot[bot] in https://github.com/srbhr/Resume-Matcher/pull/996
* chore(deps): bump cryptography from 48.0.1 to 50.0.0 in /apps/backend by @dependabot[bot] in https://github.com/srbhr/Resume-Matcher/pull/910
* build(deps): bump sharp from 0.35.3 to 0.35.4 in /apps/frontend by @dependabot[bot] in https://github.com/srbhr/Resume-Matcher/pull/998
* build(deps-dev): bump js-yaml from 4.3.0 to 4.3.2 in /apps/frontend by @dependabot[bot] in https://github.com/srbhr/Resume-Matcher/pull/997
* chore(deps): bump dompurify from 3.4.12 to 3.4.13 in /apps/frontend by @dependabot[bot] in https://github.com/srbhr/Resume-Matcher/pull/915
* build(deps): bump next from 16.2.12 to 16.3.3 in /apps/frontend by @dependabot[bot] in https://github.com/srbhr/Resume-Matcher/pull/999
* fix(builder): allow reordering items in Projects and custom sections by @Bdysj in https://github.com/srbhr/Resume-Matcher/pull/1001
* feat(builder): allow reordering description points within an entry by @srbhr in https://github.com/srbhr/Resume-Match
1.3 Crescendolls2026-09-06
Resume Matcher 1.3 — Crescendolls
Crescendolls brings resume creation, tailoring, application tracking, and interview preparation into one workflow. It also strengthens the steps you depend on most: uploading a resume, saving your work, applying AI suggestions, and exporting a PDF.
Named after Crescendolls from Daft Punk’s Discovery, this release continues the v1 series after 1.2 Nightvision.
What’s new
Build a resume from the beginning
The guided resume wizard helps you turn your experience into a resume through an adaptive conversation. Save a draft, return to it later, and continue refining the result in the builder.
Follow the application through
The application tracker gives you a board for organizing jobs, recording notes, and keeping applications connected to tailored resumes. Interview preparation helps you prepare for the next step using your resume and the job description.
More control over your resume and AI
- Three new templates: LaTeX, Clean, and Vivid.
- More language support: French and Korean interfaces and improved Chinese, Japanese, and Korean PDF rendering.
- More AI options: Groq, Azure AI Foundry, and OpenAI-compatible servers, reasoning controls, and custom prompts for cover letters and outreach messages.
- Clearer tailoring feedback: expanded score breakdowns and improved job-description skill alignment across more resume sections.
More dependable, from upload to export
This release includes a reviewed reliability update spanning 17 pull requests and 44 tracked issues.
- Recover interrupted uploads. Retain the identity of a saved resume when its response is interrupted, and handle empty or invalid documents more clearly.
- Keep useful work. Improve draft recovery, save behavior, preview confirmation, and handling of partial AI results.
- Handle failures consistently. Bound AI operation times and retries, improve malformed-response handling, and provide clearer error messages.
- Preserve resume content. Strengthen checks around original information and recognize Latin skill names beside Chinese, Japanese, and Korean text.
- Improve PDF reliability. Handle concurrent exports, cancellation, and browser shutdown while retaining ownership of resources still in use.
The reliability stack preserves 124 individual commits. Its review evidence and file map are included in the repository.
Security update
This release includes the security bump merged in PR #993: NanoID 3.3.16 → 3.3.18.
The update addresses CVE-2026-67213, where a custom ID generator could loop indefinitely when asked for a zero-length ID.
Upgrading from 1.2
Stop the application and back up the complete persistent data directory before upgrading. This release moves storage to SQLite and stores provider API keys encrypted at rest. On startup, it imports a legacy TinyDB database when SQLite is empty and migrates legacy saved keys.
Keep the database, config.json, uploaded files, and the hidden .secretkey** together. Existing encrypted keys need their original secret to be decrypted. Retain your pre-upgrade backup if you need to return to 1.2; the migrated SQLite database cannot replace the old version’s database.
For source installs, use Node.js 22+ and Python 3.13+, then reinstall dependencies following the README. For Docker upgrades, preserve your existing data volume when recreating the container.
Docker images
Release tag: v1.3.0. Versioned images are available for Linux amd64 and arm64 on GHCR and Docker Hub:
``bash
docker pull ghcr.io/srbhr/resume-matcher:1.3.0
or
docker pull srbhr/resume-matcher:1.3.0
`
The image tags are 1.3.0 and 1.3. Release-tag publication succeeded, and both tags were verified publicly on both registries.
Verification
The release preparation passed 1,067 backend tests and 500 frontend tests, plus lint, type checking, and a production build. The suite includes real Chromium controls. Two optional paid-provider evaluations were excluded.
Thank you
Thank you to everyone who contributed code, reported bugs, tested fixes, and reviewed this release. The complete contribution history and first-time contributors are listed below.
Contributors to this release: @AbdullahFageeh, @chadidi, @Dessalines39394, @Frankli9986, @gingeekrishna, @godson-plt, @GwIhViEte, @icemc, @lbagnascom, @LongD1stanceRunner, @luochen211, @moduvoice, @mohdazmi03, @MRDGH2821, @NightingaleCen, @RedMoonnn, @sahayagodson, @spsilu, @srbhr, @thmsmtylr, @wwlwgo.
Automation contributions are credited to @dependabot and @oz-agent.
<details>
<summary>All changes and first-time contributors</summary>
What's Changed
* chore(deps): bump litellm from 1.81.8 to 1.83.0 in /apps/backend by @dependabot[bot] in https://github.com/srbhr/Resume-Matcher/pull/735
* chore(deps): bump vite from 7.3.1 to 7.3.2 in /apps/frontend by @dependabot[bot] in https://github.com/srbhr/Resume-Matcher/pull/738
* chore: extract portable docs and remove unused AI tooling by @srbhr in https://github.com/srbhr/Resume-Matcher/pull/741
* chore(deps-dev): bump brace-expansion transitive deps by @srbhr in https://github.com/srbhr/Resume-Matcher/pull/744
* frontend ui audit follow-up: perf, a11y, tokens, copy, polish by @srbhr in https://github.com/srbhr/Resume-Matcher/pull/742
* fix(docker): make frontend accessible for PDF generation by @chadidi in https://github.com/srbhr/Resume-Matcher/pull/724
* chore(deps): bump next from 16.1.7 to 16.2.3 in /apps/frontend by @dependabot[bot] in https://github.com/srbhr/Resume-Matcher/pull/748
* chore(deps): bump python-multipart from 0.0.22 to 0.0.26 in /apps/backend by @dependabot[bot] in https://github.com/srbhr/Resume-Matcher/pull/752
* chore(deps): bump dompurify from 3.3.3 to 3.4.0 in /apps/frontend by @dependabot[bot] in https://github.com/srbhr/Resume-Matcher/pull/753
* Bun compatibility + PyProject script by @MRDGH2821 in https://github.com/srbhr/Resume-Matcher/pull/755
* Bun compat + uv run app` script + LiteLLM reasoning hardening (#747, #751 partial) by @srbhr in https://github.com/srbhr/Resume-Matcher/pull/756
Dev fixes: Settings error overflow, openaicompatible provider, custom feature prompts (#754, #751, #749) by @srbhr in https://github.com/srbhr/Resume-Matcher/pull/757
* fix(frontend): remove invalid turbopackUseSystemTlsCerts flag to unblock Docker Publish by @srbhr in https://github.com/srbhr/Resume-Matcher/pull/758
* fix(llm): conditionally pass temperature based on model capability by @Frankli9986 in https://github.com/srbhr/Resume-Matcher/pull/773
* test(backend): cover config settings flows by @AbdullahFageeh in https://github.com/srbhr/Resume-Matcher/pull/771
* fix(parser): prevent JSON truncation by clamping tokens to model limit by @mohdazmi03 in https://github.com/srbhr/Resume-Matcher/pull/770
* chore(deps): bump litellm from 1.83.0 to 1.83.7 in /apps/backend by @dependabot[bot] in https://github.com/srbhr/Resume-Matcher/pull/764
* chore(deps): bump python-dotenv from 1.2.1 to 1.2.2 in /apps/backend by @dependabot[bot] in https://github.com/srbhr/Resume-Matcher/pull/761
* fix(llm): resolve DeepSeek 500 / Failed to analyze resume by @Frankli9986 in https://github.com/srbhr/Resume-Matcher/pull/775
* fix(deps): pin python-dotenv to 1.0.1 to satisfy litellm (#779) by @NightingaleCen in https://github.com/srbhr/Resume-Matcher/pull/780
Remove repeated def for appears_truncated by @lbagnascom in https://github.com/srbhr/Resume-Matcher/pull/787
* [codex] Add verifier-led resume tailoring loop by @srbhr in https://github.com/srbhr/Resume-Matcher/pull/789
* fix(frontend): show timeout-specific error instead of mis
1.2 Nightvision2026-04-02
Resume Matcher 1.2(代号 Nightvision)发布。核心是新增基于 Diff 的简历改进引擎,能直观看出每轮修改的差异;同时强化了简历定制(tailor)流水线,修复了截断 JSON 需重试、自定义板块字段类型不规范等后端问题。部署上改为单一公开端口并调整 API 路由,Docker 构建失败、优雅停机和 CORS 自动加载也一并修复,本地起服务比以前省事。卡在「处理中」状态的简历现在可以重试或删除。前端方面,仪表盘卡片按标题显示字母标识与配色,经历和教育条目支持拖拽排序,PDF 下载文件名会带上姓名、公司等信息;另修正了失效的 deepseek 模型 ID,升级多项依赖并校正文档与代码不一致之处。
1.1 Voyager2026-02-10
Resume Matcher 1.1「Voyager」发布。这版重点在三块:Windows 上 asyncio 事件循环的兼容问题得到修复,Docker 配置与 uvicorn 启动命令改用 BACKENDPORT 变量,跑起来更稳;LLM 接入更健壮,apibase 会统一规范化,响应解析加了防截断与失败重试,设置页的 OpenRouter 连接测试也改为读取表单当前值。提示词结构大改,新增提示词选择与配置,简历打磨引入多轮精修,并加入防编造内容机制和改动对比视图。简历编辑器支持拖拽分区,工作经历、项目等模块可单独 AI 重新生成。多语言补齐中文、葡萄牙语,PDF 导出的构建问题一并修掉,前端依赖也做了升级。
v1.0.02026-01-06
Resume Matcher v1.0.0 发布,这是项目的首个稳定版。核心能力包括:针对目标岗位定制简历内容、内置多套 ATS 友好模板、可接入多家 AI 服务商,以及开箱即用的 Docker 部署支持。完整功能清单见项目 README 的 Key Features 一节。相较此前的 0.1.2 旧版,本次重写后整体趋于稳定,适合直接用于日常简历改写与匹配度检查。
0.1.2-old-resume-matcher2025-05-26
这是旧版 Resume Matcher 代码的存档发布(2025 年 5 月 26 日),供想基于老版本继续开发的人取用,使用时请切到 old-resume-matcher-2025 分支。
存档前的最后一批改动集中在几块:Streamlit 界面推出交互式 v1.0,新增侧边栏、favicon 与下拉式简历选择(替代滑块),并完成落地页、信息区与页脚的 UI 调整;LinkedIn 转 PDF 的逻辑得到优化,修复了 Next.js 在 TS 文件报错时抛异常的问题;Web 应用支持将简历与职位描述保存到文件系统,并迁移到独立的 web-app 分支。此外为 Windows(非 WSL)用户补了启动脚本,新增 Colab 使用方式,文档补上了贡献指南与安全策略,也加了一句不需 API Key 的说明。
依赖方面,Streamlit 升到 1.30.0,FastAPI、pypdf、pyarrow、Pillow、fonttools、jinja2、python-multipart 等也同步更新,Next.js 从 13.4.19 升到 13.5.3。
v0.1.1-alpha预发布2023-09-25
本次 v0.1.1-alpha 更新主要围绕 Web 应用与若干稳定性修复展开。
最大的变化是新增了一套基于 Next.js 与 FastAPI 的全栈 Web 界面,取代原先纯 Streamlit 的交互方式,部署后可直接在浏览器中完成简历匹配流程。同时修复了 Web 应用在 WSL 环境下 shell source 找不到的问题。
核心功能方面:解决 Cohere 与 Qdrant 使用 URL 形式 API key 时报错的问题(通过锁定 qdrant-client 最低版本并移除其循环依赖);调整 Docker 镜像的工作目录,修复 find_path() 报错;新增将 LinkedIn 职位链接转为 PDF 的功能。
其余为文档链接、README 演示动图、依赖清单与界面文案等零散调整。
v0.1.0-alpha.1 (20 Aug 23) First Alpha Release预发布2023-08-20
v0.1.0-alpha.1 是 Resume-Matcher 的首个 Alpha 版本。核心能力已跑通:简历解析、语义相似度计算,并接入了 Cohere 与 Qdrant 实现向量检索匹配。解析逻辑做了调整,仅在存在单份简历时才触发解析流程。此外修复了 NLTK 报错与部署问题,补充了 requirements.txt,Streamlit 应用也同步更新,同时批量升级了一批依赖以修复已知安全漏洞。后续计划推出完整 UI 和 React Dashboard。