memory.talk
跑 code agent 的工作台,记忆是它的副产物。
memory.talk v5 有三个顶层对象:
- work —— 做事。一件事是一个 work,复杂的事是一棵 work 树;每个 work 里盛放若干个现场(worklet):Claude Code / Codex / Kimi 工作单元、终端、网页,按 URI 打开,由对应的 work server 建出来,活得比连接久。
- metas —— 认知。一个分层的 git 仓库:
origin(外部来的原文,只读)/issue(问题 + 立场 + 论证,IBIS)/card(争完的事实,维基式词条),还可以用一份 YAML schema 加自己的层。每个动作一个[层]提交,跨层被守卫拒绝;目录下的manager.json把变动打给某个 work 的收件箱。 - user —— 人。注册的实体,和 work 平级;work 谁建的、谁在动,metas 的提交谁做的。不做权限:一个实例给一个团队用。
没有数据库、没有索引:认知层在 git 里,work / user 的记录走可换的存储 provider(本地文件系统或 SQLite,将来 S3 / MySQL)。
设计文档见 docs/designs/v5/;数据结构 docs/structure/v5/;HTTP API docs/api/v5/;CLI docs/cli/v5/。
安装
pip install memorytalk
memory.talk server start # 本地 API:http://127.0.0.1:8000/docs
需要 git 和 tmux(终端 / agent 现场由 tmuxd 建在 tmux 里;浏览器里的终端窗口是它自带的 ttyd,Linux 上 wheel 里带,不用另装)。
用起来
memory.talk user add alice # 注册(一次)
export MEMORY_TALK_USER=alice
W=$(memory.talk work create --goal '把配置改成环境变量' --json | jq -r .id)
memory.talk work attach $W codex:///home/alice/memory.talk # 在 work 里开一个 Codex 工作单元
memory.talk work recall $W # 开工注入:card 目录
memory.talk meta write issue memory.talk/配置/该走文件还是环境变量 --field question='配置该走文件还是环境变量?'
memory.talk meta act issue position memory.talk/配置/该走文件还是环境变量 --field claim='只用环境变量'
memory.talk meta act issue decide memory.talk/配置/该走文件还是环境变量 \
--field position=p1 --field card=memory.talk/配置/配置只来自环境变量
memory.talk work set $W --status archived
存储
~/.memory.talk/
├── metas/ 分层 git 仓库:layer/origin、layer/issue、layer/card(+ 用户层)、stack
├── works.db work 的现在(sqlite):works / work_columns / worklets / inbox
├── worktrace.db work 的经过(sqlite):spans / points / rounds
├── users/ user 档案(MEMORY_TALK_STORE=fs 时)
└── auth/ 登录态(MEMORY_TALK_STORE=fs 时)
work 固定是这两个 sqlite,路径可用 MEMORY_TALK_WORKS_DB / MEMORY_TALK_WORKTRACE_DB 改;MEMORY_TALK_STORE 只管 users / auth(fs 是上面的 users/ / auth/,sqlite 进 memory.sqlite);metas 永远是 git。全部环境变量见 docs/structure/v5/filesystem.md。
开发
pip install -e ".[dev]"
pytest # 每个场景在 fs 和 sqlite 两种 users / auth store 下各跑一遍(work 两种下都是 works.db + worktrace.db)
cd memorytalk/frontend && npm ci && npm run dev # 前端(Vite + React;工作台 / 工作单元 / 认知库)
发布前先 cd memorytalk/frontend && npm run build(产物 dist/ 随 wheel 分发),再 python -m build && twine upload dist/*。
布局
memorytalk/ ← Python 包(pip: memorytalk;命令 memory.talk)
├── cli/ ← 命令行(本地 API 的客户端):server / work / user / meta 各一个文件
├── backend/ ← 服务:main.py config.py gateway.py
│ ├── models/ services/ controllers/ ← 三层
│ ├── providers/ ← 存储介质:FileSystemProvider(LocalFS)/ DatabaseProvider(SQLite)
│ ├── layers/ ← 内置 layer:origin / issue / card
│ └── work_servers/ ← 每个协议一个 work server:bash / claude / codex / kimi / http / default
└── frontend/ ← Vite + React 工作台(工作导航、工作单元、认知库、设置)
tests/ ← 按场景组织
docs/ ← designs / structure / api / cli
License
1 maintainer
Metadata
1 maintainer
Release files for memorytalk 5.2.0
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | Provenance | |
|---|---|---|---|---|
| memorytalk-5.2.0.tar.gz | 2.9 MB | Without provenance | Details |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Reset |
|---|---|---|---|---|
| memorytalk-5.2.0-py3-none-any.whl | Python 3 | none | any | Details |
Total release size: 5.2 MB
Release files / memorytalk-5.2.0.tar.gz
| Download URL | memorytalk-5.2.0.tar.gz |
|---|---|
| Size | 2.9 MB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
49cf3f1ef569871795042ceae32514bffa26d5cd0d462f48f6d72ce5b8224a05
|
|
BLAKE2b-256 checksum How to use checksums |
422b35f3b04bf9be4bee00542e6ed38667d74a7fc89e3a2522044abd090541cc
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/7.0.0 CPython/3.11.2
|
Release files / memorytalk-5.2.0-py3-none-any.whl
| Download URL | memorytalk-5.2.0-py3-none-any.whl |
|---|---|
| Size | 2.3 MB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
fee4fac272d88be8534ca10847557589dec63e48c3f760e0aca88d6aa97e4d42
|
|
BLAKE2b-256 checksum How to use checksums |
4c15d320ce25f7cc61ca2b834c00ec6cefccc71deec81d4a4839ddda3e958733
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/7.0.0 CPython/3.11.2
|
Security for memorytalk 5.2.0
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
No provenance found
None of the files in this release were uploaded with provenance metadata.
A note on package security
PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.
