할 수 있는 일
Provenance tree for coding agents: where you are, why, and what was already decided
제작자가 쓴 원문
- 연결 방식 내 컴퓨터에서 실행
- 계정 또는 키 키 필요 없음
- 분류 개발자 도구
함께 쓰기
이런 작업을 AI에게 가르치는 Skill, 이것과 함께 쓸 수 있습니다
andrej-karpathy-skills
A single CLAUDE.md file to improve Claude Code behavior, derived from Andrej Karpathy's observations on LLM coding pitfalls.
graphify
Turn any codebase, with its docs, SQL schemas, configs, and PDFs, into a queryable knowledge graph. A /graphify skill for Claude Code, Cursor, Codex, and Gemini CLI: local deterministic AST parsing, every edge explained, no vector store.
이 순위인 이유
목록은 사용 점수(100점 만점)로 정렬되며 구성은 다음과 같습니다
| 인기 | 4/80 | GitHub 별 또는 주간 다운로드: 큰 쪽 60, 작은 쪽 20 |
|---|---|---|
| 최신성 | 12/12 | 30일 안에 릴리스나 코드 변경이 있으면 만점 |
| 응답 중 | 0/8 | 마지막 점검 때 클라우드 엔드포인트가 응답 |
| 사용 점수 | 16/100 |
품질: 기본 점검 3/3개 통과(표시만, 정렬에는 안 씀) · 상승세: 데이터 없음 (급상승 탭에서만 사용)
설치
쓰고 있는 AI 앱을 고른 다음 설정을 복사하세요.
검사 3/3 통과 · 내 컴퓨터에서 실행돼요
먼저: 어떤 AI를 쓰시나요?
쓰는 앱을 고르면 아래에 단계가 나옵니다.
Claude는 아직 vivac에 연결할 수 없어요
제작자가 아직 MCP 레지스트리에 설치 방법을 올리지 않았어요
ChatGPT는 아직 vivac에 연결할 수 없어요
제작자가 아직 MCP 레지스트리에 설치 방법을 올리지 않았어요
대신 이렇게 할 수 있어요
B · 개발자 도구에서 ChatGPT에 연결되는 다른 서버
같은 분류에서 많이 쓰는 것 중 키 없이 주소만으로 연결되는 것을 골랐어요
- Sentry✓ ChatGPT
- Compiler Explorer✓ ChatGPT
- GitDiagram✓ ChatGPT
Claude Code는 아직 vivac에 연결할 수 없어요
제작자가 아직 MCP 레지스트리에 설치 방법을 올리지 않았어요
Cursor는 아직 vivac에 연결할 수 없어요
제작자가 아직 MCP 레지스트리에 설치 방법을 올리지 않았어요
VS Code는 아직 vivac에 연결할 수 없어요
제작자가 아직 MCP 레지스트리에 설치 방법을 올리지 않았어요
한눈에 보기
어떤 앱에서 쓸 수 있나요
이 서버의 연결 방식과 각 앱의 공식 문서를 바탕으로 했어요. 2026년 11월 10일 확인.
내 컴퓨터에서 실행하는 앱
GooseLM StudioMsty StudioCherry StudioVisual StudioWindsurf (Devin Desktop)ZedKiroClineContinueJetBrains AI AssistantCodex CLIGemini CLIOpenCodenodecodeWarpLibreChatRaycast
할 수 있는 일
내 컴퓨터에서 실행돼요. 설치하면 AI 앱에 도구 목록이 나타나요.
기본 점검i
누구나 볼 수 있는 정보를 점검했고, 코드는 읽지 않았어요. 해당하지 않는 항목은 개수에서 뺐어요. · 2026년 11월 20일 점검
- ✓ 통과
제작자와 저장소 일치
레지스트리 이름이 GitHub 계정 JAAvila-Of 기준으로 확인되었고, 이 계정이 저장소를 소유하고 있어요
- ✓ 통과
계속 관리되고 있음
코드 마지막 변경: 12일 전
- ✓ 통과
라이선스 있음
Apache-2.0 사용
- – 해당 없음
엔드포인트 응답
내 컴퓨터에서 실행되는 서버라 인터넷 엔드포인트가 없어요
- – 해당 없음
OAuth로 로그인
내 컴퓨터에서 실행되며, 대신 로컬 설정을 사용해요
- … 아직 점검 안 함
도구 설명에 숨겨진 지시 없음
내 컴퓨터에서 실행돼요. 도구 목록은 인터넷 엔드포인트가 있는 서버에서만 읽을 수 있어요
제작자: 재확인 요청 또는 잘못된 정보 신고
제작자용: README에 배지 달기
제작자라면 이 코드를 README에 붙여 넣으세요. 배지는 이 페이지로 연결됩니다.
[](https://mcp.popnx.com/en/servers/jaavila-of-vivac)토론
질문하고, 팁을 나누고, 안 되는 점을 알려 주세요. "도움이 됨"이 많은 순서예요.
사용자 리뷰
아직 리뷰가 없어요
개발자 도구 카테고리의 다른 서버와 비교i
- 이 카테고리의 다른 서버 6209개 중 69%보다 스타가 많아요
- 이 카테고리의 다른 서버 6209개 중 64%보다 최근에 레지스트리에 공개됐어요
- 기본 점검 결과가 이 카테고리의 다른 서버 6209개 중 99%와 같거나 더 좋아요
레지스트리에서 이 카테고리의 활성 서버 전체와 비교했어요. 스타는 게시자가 저장소를 소유한 경우에만 세요.
자주 묻는 질문
vivac, 안전한가요?
해당하는 기본 점검 3개 중 3개를 통과했어요. 코드는 읽지 않았어요. 키나 토큰을 넣기 전에 서버가 어떤 권한을 요구하는지 확인하고, 꼭 필요한 권한만 있는 키를 쓰세요.
어떤 키나 코드가 필요한가요?
레지스트리에 필수 키가 나와 있지 않아요. 다만 처음 쓸 때 로그인을 요구하는 서버도 있어요.
내 컴퓨터에서 실행되나요, 클라우드에서 실행되나요?
내 컴퓨터에서 실행돼요. AI 앱이 필요할 때 실행하고, 데이터는 제작자의 서버를 거치지 않아요(서버 자체가 온라인 서비스를 호출하는 경우는 제외).
아직 관리되고 있나요?
GitHub 저장소는 마지막으로 12일 전 변경됐고, 최신 버전(0.22.1)은 16일 전 레지스트리에 올라왔어요.
어떤 앱에서 쓸 수 있나요?
Claude Desktop, Claude Code, Cursor, VS Code처럼 MCP를 지원하는 앱에서 쓸 수 있어요. 내 컴퓨터에 Rust도 있어야 해요.
대신 쓸 수 있는 건 뭐가 있나요?
같은 카테고리에는: GitDiagram, Appllama, Cloudflare MCP, Omnara, Repowise
비슷하게 동작하는 다른 것
같은 카테고리, 다른 제작자, 명령 확인 완료. 공식이 먼저, 그다음 스타 순.
카드 보는 법:↓ 주간 다운로드i★ GitHub 스타i세 기둥: 리뷰 · 점검 · 사용 정도i업데이트 = 최신 버전을 레지스트리에 공개한 날i
GitDiagram
Architecture diagrams, explanations and Mermaid source for public GitHub repositories.
Appllama
Study screens, flows and paywalls from top-earning iOS apps, then build from what wins.
Omnara
Deploy durable AI agents with Omnara, an open-source managed-agent platform.
Repowise
Codebase intelligence for AI coding agents: graph, git history, docs, decisions, code health.
| 이 항목 vivac | GitDiagram | Appllama | Cloudflare MCP | Omnara | Repowise | |
|---|---|---|---|---|---|---|
| 실행 위치 | 컴퓨터에 설치 필요 | 클라우드에서 동작, 설치 불필요 | 클라우드에서 동작, 설치 불필요 | 클라우드에서 동작, 설치 불필요 | 클라우드에서 동작, 설치 불필요 | 클라우드 또는 컴퓨터 설치 모두 가능 |
| 키 필요 | 아니요 | 아니요 | 아니요 | 아니요 | 아니요 | 아니요 |
| Claude에서 +로 추가 | 아니요 | 아니요 | 아니요 | 예 | 아니요 | 아니요 |
| 기본 점검 | 3/3 | 4/4 | 4/4 | 4/4 | 4/4 | 5/5 |
| 라이선스 | Apache-2.0 | MIT | MIT | Apache-2.0 | Apache-2.0 | AGPL-3.0 |
| 가격 | 등록 정보에 비용 없음 | 등록 정보에 비용 없음 | 등록 정보에 비용 없음 | 등록 정보에 비용 없음 | 등록 정보에 비용 없음 | 등록 정보에 비용 없음 |
제작자의 README
README 열기
제작자의 저장소에서 쓰인 그대로 가져왔어요. PopNx MCP가 검토하거나 보증한 내용이 아니에요. HTML, 이미지, 스크립트는 제거했어요. 링크를 누르면 이 사이트를 벗어나요.
A tree where every node knows which node it was born from.
*So that months later something can still answer “why are we here?”*
$ vivac why 4
Why we are here -> t4
g1 Ship the 2.0 API the first customer is waiting on it (4 open / 1 closed below) | v t2 Replace the cache adapter the session bug traces back to it (3 open / 1 closed below) | v t4 No test for expiry [closed] no way to reproduce the session bug ! the corpus run is what settled it = reproduced: sessions expire at 300s, not 3600
^^^ you are here
In parallel, still open (3): t3 Rate limiting is undecided d5 Retry policy: three tries, then fail loudly t6 Migrate the callers
t2 does not close until these close (1): t6 Migrate the callers
[What you get](#what-you-get) · [See it](#see-it) · [Install](#install) · [First five minutes](#the-first-five-minutes) · [Why one map](#one-map) · [Bring a project in](docs/MIGRATING.md)
Built for one person's own work
I run eight projects in parallel — open source, work and my own — and five of them are large. Every time I came back to one I had to piece together what had been decided in it, and more than once I watched the agent change something we had already settled, because both of us had forgotten we had. What I had against that was manual: ending each stretch of work by asking the model for *a safepoint, so we can pick this up later*, or keeping a where_are_we.txt at the root of the project.
vivac is what replaced them, and it is still measured on those projects. That is the whole of its pedigree, and it shows in what got built: every mechanism here came out of a defect that had already cost me days, and every number on this page came off a real tree rather than a benchmark written to make a README look good.
The tree this project keeps of itself, 23 days in: 695 nodes, 315 of them closed, 176 standing decisions, 16 levels deep.
Three of those defects, and what each one turned into:
- A run marked `DONE` with its findings still open — and 26 days before anybody noticed. Now
vivac donerefuses, and says what is missing. - A reading list of 109 open fronts across 224 lines, with the one touched yesterday at the bottom. Now
openanswers *what is waiting on you*, in that order, and stops at ten. - Three claims shipped to crates.io that the binary beside them contradicted. Now a test runs every command this page shows and holds its lists against
--help.
What I would not give up now is no single feature. It is looking at the tree of what was decided and why; picking a project back up with whichever model is at hand and having it know what the last one settled; and knowing where the work stands at any moment. In my experience that is worth more than anything else here, and I hope it serves you as well.
Issues are open, and I want to hear where it fails you. Pull requests are not open yet; CONTRIBUTING.md says why.
The problem
When you develop with an agentic AI, work spawns more work. Three hops in, you have lost the thread of what you originally set out to do.
It is not a memory problem: usually everything is written down. It is a provenance problem. What is written does not say what it was born *from*, and without that edge there is no way to reconstruct why you are where you are.
Measured on a real compiler: the path between the goal and the day's work
was six levels deep, spread across a chronologically ordered 8,853-line
tracker, 52 planning documents and 21 issues. The structure was temporal,
which is exactly the opposite of provenance.
Logbooks, decision records, issue trackers and session memory for agents all store the node. Where one of them stores the edge as well, it is a link somebody has to remember to add — so it is missing on exactly the node where nobody thought it would matter. Where it sits goes through them category by category, and says where each one is better than this.
What you get
The agent starts oriented, and nobody has to ask it to. A hook runs vivac session start when a session opens, so the first thing in its context is where you are, what has already been decided, and what not to touch:
$ vivac brief
vivac · project: demo · lane: main · 2026-09-21
GOAL g1 Ship the 2.0 API | |-- t2 Replace the cache adapter | why: the session bug traces back to it | `-- t6 Migrate the callers <== HERE why: the old adapter had a different signature
STANDING DECISIONS d5 Retry policy: three tries, then fail loudly
LAST VIVAC v5 · push · 2026-09-21 · bcdba21 you were about to: Migrate the callers
143 tokens · depth 3 · 0 parked
In tokens, that is the whole argument. A project keeping its state in three places was asked to pick up where it left off. A hand-written plan answered in 9,252 tokens. A memory system answered *“maybe”* in about 12,720, depending on which of two names for the project it resolved. The tree answered in 100. The brief carries a token budget because a context window is the one resource every session spends.
Nothing closes over what is still open. The one operation in the model that rejects, and it earns it: a run marked done over open findings took 26 days to be spotted once.
$ vivac done 2
t2 CANNOT close: 1 open closure condition(s)
t6 Migrate the callers
A run closes with its findings, not with its report. Closing it anyway leaves a trace: vivac done 2 --force
Every decision keeps what it turned down, and what it was judged against. --alternative holds the option rejected, --supersedes links a reversal to what it reverses, and --against records the rule or pillar that decided it — so a decision can be argued with a year later instead of guessed at. This project's own tree carries 176 of them.
An assumption that falls does not take its children with it. abandon marks the premise refuted and everything under it goes with it, except what you rescue — and what is rescued still hangs where it was born, because being born somewhere is not undone by that place turning out to be wrong.
What it does not promise
It saves tokens, not all of them. Picking a project up took the tree 100 tokens where a hand-written plan took 9,252, and that difference comes back every time a session opens. The agent still reads files, still reasons and still explains itself, and none of that gets cheaper.
It does not make sure nothing is ever forgotten. No memory system can, this one included, because every one of them still runs on the model's judgement: *should I save this? is this a finding? do I need to read the tree again before I answer?* The best skill in the world, a hundred subagents or a hundred daemons move where that judgement happens, and none of them removes it. vivac hangs capture off the seams of the work rather than off that judgement, and still measures where it slips.
What is left over after that is why the tree is built to be looked at.
See it
vivac web
The agent writes the tree, and you can read it at any moment without asking the agent anything. vivac web opens every project on this machine in a browser: which one moved and which has been sitting still, what changed in one while you were away, a node's whole lineage, and the whole tree. Its Agents page compares native versions across harnesses and applies an explicitly reviewed synchronization plan through the same operations as vivac agents sync. It is also where the Artifacts page connects retained deliveries to their producer and decisions referencing exact revisions. Checks, capture and restore use the artifact engine; downloads verify the selected revision before sending it as an attachment. Static HTML, CSS and JavaScript bundles retain all declared members as one revision and restore into a new directory. HTML revisions and bundles can open in a restricted browser preview separate from the Vivac session. See Artifact custody. This is where you see the state of each item for yourself, and where you catch what the agent let pass.
Four example projects, written by real vivac commands with tools/web-screenshots.py, which takes these pictures again whenever the pages change.
A server you start and that dies when you close it, bound to 127.0.0.1, reachable through a one-time key it prints. It has no functions of its own: anything a page needs is built on the command line first. → What the web is, and is not
One map
**vivac was not