cerebro-memory 0.2.0__tar.gz
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- cerebro_memory-0.2.0/LICENSE +21 -0
- cerebro_memory-0.2.0/PKG-INFO +264 -0
- cerebro_memory-0.2.0/README.md +237 -0
- cerebro_memory-0.2.0/pyproject.toml +62 -0
- cerebro_memory-0.2.0/pyproject.toml.orig +52 -0
- cerebro_memory-0.2.0/src/cerebro_mcp/__init__.py +3 -0
- cerebro_memory-0.2.0/src/cerebro_mcp/__main__.py +3 -0
- cerebro_memory-0.2.0/src/cerebro_mcp/cli.py +363 -0
- cerebro_memory-0.2.0/src/cerebro_mcp/config.py +248 -0
- cerebro_memory-0.2.0/src/cerebro_mcp/contradictions.py +499 -0
- cerebro_memory-0.2.0/src/cerebro_mcp/demo.py +102 -0
- cerebro_memory-0.2.0/src/cerebro_mcp/demo_vault/README.md +48 -0
- cerebro_memory-0.2.0/src/cerebro_mcp/demo_vault/agent-memory/streak-app/memory/MEMORY.md +4 -0
- cerebro_memory-0.2.0/src/cerebro_mcp/demo_vault/agent-memory/streak-app/memory/monetization.md +16 -0
- cerebro_memory-0.2.0/src/cerebro_mcp/demo_vault/agent-memory/streak-app/memory/project-overview.md +19 -0
- cerebro_memory-0.2.0/src/cerebro_mcp/demo_vault/agent-memory/streak-app/memory/release-process.md +12 -0
- cerebro_memory-0.2.0/src/cerebro_mcp/demo_vault/agent-memory/streak-app/memory/user-preferences.md +15 -0
- cerebro_memory-0.2.0/src/cerebro_mcp/demo_vault/agent-memory/streak-web/memory/MEMORY.md +3 -0
- cerebro_memory-0.2.0/src/cerebro_mcp/demo_vault/agent-memory/streak-web/memory/feedback-copy.md +14 -0
- cerebro_memory-0.2.0/src/cerebro_mcp/demo_vault/agent-memory/streak-web/memory/pricing.md +16 -0
- cerebro_memory-0.2.0/src/cerebro_mcp/demo_vault/agent-memory/streak-web/memory/site-stack.md +15 -0
- cerebro_memory-0.2.0/src/cerebro_mcp/demo_vault/cerebro.toml +35 -0
- cerebro_memory-0.2.0/src/cerebro_mcp/demo_vault/notes/Clippings/Why 7-day trials convert better.md +19 -0
- cerebro_memory-0.2.0/src/cerebro_mcp/demo_vault/notes/Daily/2026-11-03.md +14 -0
- cerebro_memory-0.2.0/src/cerebro_mcp/demo_vault/notes/Daily/2026-11-17.md +16 -0
- cerebro_memory-0.2.0/src/cerebro_mcp/demo_vault/notes/Decisions/2026-09-14 Precio y anuncios.md +29 -0
- cerebro_memory-0.2.0/src/cerebro_mcp/demo_vault/notes/Dev/Payments setup.md +24 -0
- cerebro_memory-0.2.0/src/cerebro_mcp/demo_vault/notes/Dev/Sync architecture.md +19 -0
- cerebro_memory-0.2.0/src/cerebro_mcp/demo_vault/notes/Ideas/Monetization brainstorm.md +20 -0
- cerebro_memory-0.2.0/src/cerebro_mcp/demo_vault/notes/Ideas/Streak widgets.md +13 -0
- cerebro_memory-0.2.0/src/cerebro_mcp/demo_vault/notes/Journal/2026-10-30.md +13 -0
- cerebro_memory-0.2.0/src/cerebro_mcp/demo_vault/notes/Marketing/Landing copy.md +22 -0
- cerebro_memory-0.2.0/src/cerebro_mcp/demo_vault/notes/Meetings/2026-10-08 Beta testers call.md +20 -0
- cerebro_memory-0.2.0/src/cerebro_mcp/demo_vault/notes/Projects/Launch plan.md +27 -0
- cerebro_memory-0.2.0/src/cerebro_mcp/demo_vault/notes/Projects/Roadmap 2026.md +17 -0
- cerebro_memory-0.2.0/src/cerebro_mcp/demo_vault/notes/Streak.md +32 -0
- cerebro_memory-0.2.0/src/cerebro_mcp/demo_vault/notes/Templates/Daily template.md +7 -0
- cerebro_memory-0.2.0/src/cerebro_mcp/demo_vault/notes/people/Marta Ruiz.md +12 -0
- cerebro_memory-0.2.0/src/cerebro_mcp/demo_vault/repo-docs/streak-app/AGENTS.md +9 -0
- cerebro_memory-0.2.0/src/cerebro_mcp/demo_vault/repo-docs/streak-app/docs/handoff-2026-11-18.md +26 -0
- cerebro_memory-0.2.0/src/cerebro_mcp/demo_vault/repo-docs/streak-app/docs/release-checklist.md +14 -0
- cerebro_memory-0.2.0/src/cerebro_mcp/demo_vault/repo-docs/streak-web/docs/pricing-page.md +20 -0
- cerebro_memory-0.2.0/src/cerebro_mcp/embeddings.py +166 -0
- cerebro_memory-0.2.0/src/cerebro_mcp/index.py +392 -0
- cerebro_memory-0.2.0/src/cerebro_mcp/notes.py +288 -0
- cerebro_memory-0.2.0/src/cerebro_mcp/query.py +386 -0
- cerebro_memory-0.2.0/src/cerebro_mcp/redact.py +161 -0
- cerebro_memory-0.2.0/src/cerebro_mcp/server.py +332 -0
- cerebro_memory-0.2.0/src/cerebro_mcp/sources.py +182 -0
- cerebro_memory-0.2.0/src/cerebro_mcp/text.py +54 -0
|
@@ -0,0 +1,21 @@
|
|
|
1
|
+
MIT License
|
|
2
|
+
|
|
3
|
+
Copyright (c) 2026 César Méndez
|
|
4
|
+
|
|
5
|
+
Permission is hereby granted, free of charge, to any person obtaining a copy
|
|
6
|
+
of this software and associated documentation files (the "Software"), to deal
|
|
7
|
+
in the Software without restriction, including without limitation the rights
|
|
8
|
+
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
|
|
9
|
+
copies of the Software, and to permit persons to whom the Software is
|
|
10
|
+
furnished to do so, subject to the following conditions:
|
|
11
|
+
|
|
12
|
+
The above copyright notice and this permission notice shall be included in all
|
|
13
|
+
copies or substantial portions of the Software.
|
|
14
|
+
|
|
15
|
+
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
|
|
16
|
+
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
|
|
17
|
+
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
|
|
18
|
+
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
|
|
19
|
+
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
|
|
20
|
+
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
|
|
21
|
+
SOFTWARE.
|
|
@@ -0,0 +1,264 @@
|
|
|
1
|
+
Metadata-Version: 2.4
|
|
2
|
+
Name: cerebro-memory
|
|
3
|
+
Version: 0.2.0
|
|
4
|
+
Summary: A local, read-only index over your Markdown notes and your AI agents' memories, served over MCP, with a tool that finds where they contradict each other.
|
|
5
|
+
Keywords: mcp,obsidian,markdown,search,embeddings,sqlite,claude
|
|
6
|
+
Author: cerebro-memory contributors
|
|
7
|
+
License-Expression: MIT
|
|
8
|
+
License-File: LICENSE
|
|
9
|
+
Classifier: Development Status :: 3 - Alpha
|
|
10
|
+
Classifier: Environment :: Console
|
|
11
|
+
Classifier: Intended Audience :: Developers
|
|
12
|
+
Classifier: Operating System :: OS Independent
|
|
13
|
+
Classifier: Programming Language :: Python :: 3
|
|
14
|
+
Classifier: Programming Language :: Python :: 3.11
|
|
15
|
+
Classifier: Programming Language :: Python :: 3.12
|
|
16
|
+
Classifier: Programming Language :: Python :: 3.13
|
|
17
|
+
Classifier: Topic :: Text Processing :: Indexing
|
|
18
|
+
Requires-Dist: fastembed>=0.9,<1
|
|
19
|
+
Requires-Dist: mcp>=2.3,<3
|
|
20
|
+
Requires-Dist: numpy>=1.26
|
|
21
|
+
Requires-Dist: platformdirs>=4
|
|
22
|
+
Requires-Dist: pydantic>=2.7
|
|
23
|
+
Requires-Dist: pyyaml>=6.0
|
|
24
|
+
Requires-Dist: sqlite-vec>=0.1.6
|
|
25
|
+
Requires-Python: >=3.11
|
|
26
|
+
Description-Content-Type: text/markdown
|
|
27
|
+
|
|
28
|
+
# cerebro-memory
|
|
29
|
+
|
|
30
|
+
**A local, read-only index over your Markdown notes *and* your AI agents' memories, served over MCP, that tells you when they contradict each other.**
|
|
31
|
+
|
|
32
|
+
<!-- demo gif -->
|
|
33
|
+
|
|
34
|
+
[Español](README.es.md)
|
|
35
|
+
|
|
36
|
+
## Why
|
|
37
|
+
|
|
38
|
+
You write a decision in your notes. Your coding agent saved an older version of
|
|
39
|
+
it in its memory three months ago. A handoff doc in another repo says a third
|
|
40
|
+
thing. Nobody notices until an agent confidently ships the old price, the old
|
|
41
|
+
launch date or the feature you already killed.
|
|
42
|
+
|
|
43
|
+
`cerebro` indexes all of it on your machine (your Obsidian vault, Claude Code
|
|
44
|
+
memories, repo docs and handoffs) and gives you, and any MCP client, one place
|
|
45
|
+
to search it and one command to ask: *where do my sources disagree?*
|
|
46
|
+
|
|
47
|
+
- **Hybrid search** that works in English and Spanish (and other languages):
|
|
48
|
+
exact names via full-text search, paraphrases via multilingual embeddings.
|
|
49
|
+
- **Contradictions**: pairs of fragments on a topic that state different
|
|
50
|
+
prices, dates, versions or yes/no answers, with both sources and dates.
|
|
51
|
+
- **Local and read-only.** Your files are never edited. Private notes never
|
|
52
|
+
leave the index. Secrets are redacted before anything is stored.
|
|
53
|
+
|
|
54
|
+
## Quickstart (30 seconds)
|
|
55
|
+
|
|
56
|
+
You need [uv](https://docs.astral.sh/uv/).
|
|
57
|
+
|
|
58
|
+
```sh
|
|
59
|
+
uvx cerebro-memory demo
|
|
60
|
+
```
|
|
61
|
+
|
|
62
|
+
That copies a small fictional vault to a temp folder, indexes it and prints the
|
|
63
|
+
contradictions it finds. The first run downloads the embedding model once;
|
|
64
|
+
after that everything runs offline.
|
|
65
|
+
|
|
66
|
+
Then point it at your own notes:
|
|
67
|
+
|
|
68
|
+
```sh
|
|
69
|
+
uv tool install cerebro-memory # installs the `cerebro` command
|
|
70
|
+
cerebro init --vault ~/Obsidian # writes cerebro.toml; adds ~/.claude/projects if it exists
|
|
71
|
+
cerebro index # incremental; re-run any time
|
|
72
|
+
cerebro search "pricing decision"
|
|
73
|
+
cerebro contradictions "launch date" --across-sources
|
|
74
|
+
```
|
|
75
|
+
|
|
76
|
+
## What it finds
|
|
77
|
+
|
|
78
|
+
Real output of `uvx cerebro-memory demo` (it prints the strongest disagreement per pair of notes; ask about one topic with `cerebro contradictions "<topic>"`):
|
|
79
|
+
|
|
80
|
+
```text
|
|
81
|
+
$ uvx cerebro-memory demo
|
|
82
|
+
Demo vault copied to /tmp/cerebro-demo-x1y2z3/demo_vault
|
|
83
|
+
Indexed 26 notes from 3 sources (notes, agent-memory, repo-docs): 2 private (kept out of every result), 1 line(s) redacted, mode hybrid.
|
|
84
|
+
|
|
85
|
+
Your notes and your agents disagree on 3 things (older -> newer):
|
|
86
|
+
|
|
87
|
+
1. price: $4.99/month -> $5.99/month; price: $35/year -> $39/year; number: 7-day -> 14 days
|
|
88
|
+
older 2026-07-02 agent-memory:streak-web/memory/pricing.md
|
|
89
|
+
- Pro monthly: $4.99/month - Pro yearly: $35/year - 7-day free trial on both plans. Why: set when the pricing page was first built, from Lucía's brai…
|
|
90
|
+
newer 2026-09-15 repo-docs:streak-web/docs/pricing-page.md
|
|
91
|
+
…Price | Free trial | |---|---|---| | Free | $0, up to 3 habits | none | | Pro monthly | $5.99/month | 14 days | | Pro yearly | $39/year | 14 days | …
|
|
92
|
+
|
|
93
|
+
2. version: 0.8.5 -> 0.9.3 / 0.9.0; date: December 1, 2026 -> 2027-01-12 / 2026-12-15
|
|
94
|
+
older 2026-08-21 agent-memory:streak-app/memory/project-overview.md
|
|
95
|
+
…SQLite file is the source of truth; sync is optional and Pro-only. - Current version: 0.8.5 (beta channel). - Launch date: December 1, 2026. - Free …
|
|
96
|
+
newer 2026-11-18 repo-docs:streak-app/docs/handoff-2026-11-18.md
|
|
97
|
+
- Current version: 0.9.3, in the beta channel since this morning. It fixes the midnight rollover in StreakClock.rollover() for UTC+9 and UTC-10. - La…
|
|
98
|
+
|
|
99
|
+
3. number: 7-day -> 14 días
|
|
100
|
+
older 2026-07-02 agent-memory:streak-web/memory/pricing.md
|
|
101
|
+
- Pro monthly: $4.99/month - Pro yearly: $35/year - 7-day free trial on both plans. Why: set when the pricing page was first built, from Lucía's brai…
|
|
102
|
+
newer 2026-11-17 notes:Daily/2026-11-17.md
|
|
103
|
+
Martes tranquilo. Revisé las conversiones del beta: con la prueba de 14 días, 4 de 11 testers activaron Pro al terminarla. Me quedo con 14 días aunqu…
|
|
104
|
+
|
|
105
|
+
Run it on your own notes: cerebro init --vault ~/path/to/notes && cerebro index
|
|
106
|
+
```
|
|
107
|
+
|
|
108
|
+
The demo vault is in [`src/cerebro_mcp/demo_vault/`](src/cerebro_mcp/demo_vault/);
|
|
109
|
+
its README lists every fact planted in it.
|
|
110
|
+
|
|
111
|
+
## Connect it to your agents
|
|
112
|
+
|
|
113
|
+
`cerebro serve` is an MCP server over stdio. Your client launches it; you never
|
|
114
|
+
run it by hand. If you installed the tool, the command is `cerebro serve`;
|
|
115
|
+
without installing, it is `uvx cerebro-memory serve`.
|
|
116
|
+
|
|
117
|
+
**Claude Code**
|
|
118
|
+
|
|
119
|
+
```sh
|
|
120
|
+
claude mcp add cerebro --scope user -- uvx cerebro-memory serve
|
|
121
|
+
```
|
|
122
|
+
|
|
123
|
+
**Codex** (`~/.codex/config.toml`)
|
|
124
|
+
|
|
125
|
+
```toml
|
|
126
|
+
[mcp_servers.cerebro]
|
|
127
|
+
command = "uvx"
|
|
128
|
+
args = ["cerebro-memory", "serve"]
|
|
129
|
+
```
|
|
130
|
+
|
|
131
|
+
**Cursor** (`~/.cursor/mcp.json`) and **Claude Desktop** (`claude_desktop_config.json`)
|
|
132
|
+
|
|
133
|
+
```json
|
|
134
|
+
{
|
|
135
|
+
"mcpServers": {
|
|
136
|
+
"cerebro": {
|
|
137
|
+
"command": "uvx",
|
|
138
|
+
"args": ["cerebro-memory", "serve"]
|
|
139
|
+
}
|
|
140
|
+
}
|
|
141
|
+
}
|
|
142
|
+
```
|
|
143
|
+
|
|
144
|
+
To use a config file other than the default, add
|
|
145
|
+
`"env": { "CEREBRO_CONFIG": "/path/to/cerebro.toml" }`. More detail, file
|
|
146
|
+
locations and Windows notes: [docs/connect.md](docs/connect.md).
|
|
147
|
+
|
|
148
|
+
### MCP tools
|
|
149
|
+
|
|
150
|
+
| Tool | What it does |
|
|
151
|
+
|---|---|
|
|
152
|
+
| `search` | Hybrid search with filters (`source`, `scope`, `type`, `since`, `until`) |
|
|
153
|
+
| `note` | Full text, frontmatter and links of one note (by id, path or title) |
|
|
154
|
+
| `links` | Outgoing and incoming `[[links]]` |
|
|
155
|
+
| `todos` | Open checkboxes `- [ ]` in the notes that match a query |
|
|
156
|
+
| `contradictions` | Pairs of fragments on a topic that state different values |
|
|
157
|
+
| `propose` | Writes a Markdown proposal into cerebro's own folder; never touches your notes |
|
|
158
|
+
|
|
159
|
+
## Configuration
|
|
160
|
+
|
|
161
|
+
`cerebro init` writes a starter file. A full example:
|
|
162
|
+
|
|
163
|
+
```toml
|
|
164
|
+
[[source]]
|
|
165
|
+
name = "notes"
|
|
166
|
+
kind = "vault" # Obsidian/Markdown: frontmatter, [[links]], checkboxes
|
|
167
|
+
path = "~/Obsidian"
|
|
168
|
+
scope = "personal"
|
|
169
|
+
exclude = ["Templates/**"]
|
|
170
|
+
|
|
171
|
+
[[source]]
|
|
172
|
+
name = "claude-memory"
|
|
173
|
+
kind = "claude-code-memory" # <path>/<project>/memory/*.md
|
|
174
|
+
path = "~/.claude/projects"
|
|
175
|
+
|
|
176
|
+
[[source]]
|
|
177
|
+
name = "handoffs"
|
|
178
|
+
kind = "glob" # any Markdown matched by the include globs
|
|
179
|
+
path = "~/code"
|
|
180
|
+
include = ["*/docs/**/*.md", "*/AGENTS.md", "*/CLAUDE.md"]
|
|
181
|
+
scope = "work"
|
|
182
|
+
|
|
183
|
+
[privacy]
|
|
184
|
+
private_folders = ["private", "people"]
|
|
185
|
+
private_frontmatter = "private"
|
|
186
|
+
redact = ["secrets", "cards"]
|
|
187
|
+
```
|
|
188
|
+
|
|
189
|
+
The config is looked up in this order: `--config PATH`, `$CEREBRO_CONFIG`,
|
|
190
|
+
`./cerebro.toml`, then your user config folder. Run `cerebro doctor` to see
|
|
191
|
+
which file was used, how many files each source has and whether the model is
|
|
192
|
+
cached.
|
|
193
|
+
|
|
194
|
+
## Privacy model
|
|
195
|
+
|
|
196
|
+
- **Read-only.** cerebro never edits, moves or deletes a source file. The only
|
|
197
|
+
thing it writes besides its index is `propose`, into its own proposals folder.
|
|
198
|
+
- **Local.** Parsing, indexing and embeddings run on your machine. The only
|
|
199
|
+
network call is the one-time model download. Only the fragments a client asks
|
|
200
|
+
for leave the process, and only to that client.
|
|
201
|
+
- **Private notes stay private.** A note in a private folder (`private/`,
|
|
202
|
+
`people/` by default) or with `private: true` is indexed by path and hash
|
|
203
|
+
only. Every query filters private notes out in SQL, so no tool can return them.
|
|
204
|
+
- **Redaction before storage.** Lines with API keys, tokens, private keys or
|
|
205
|
+
card numbers are redacted before the note is parsed. Packs: `secrets`,
|
|
206
|
+
`cards`, and `mx` (Mexican IDs and bank numbers).
|
|
207
|
+
- **Hidden scopes.** Sources in a hidden scope are indexed but only returned
|
|
208
|
+
when that scope is asked for by name.
|
|
209
|
+
- **Content is data, not instructions.** Every tool output carries a short
|
|
210
|
+
notice saying so, and web clips (`source: web`) are marked untrusted.
|
|
211
|
+
|
|
212
|
+
Details and limits: [docs/privacy.md](docs/privacy.md).
|
|
213
|
+
|
|
214
|
+
## How it works
|
|
215
|
+
|
|
216
|
+
```text
|
|
217
|
+
Markdown sources vault · agent memories · repo docs
|
|
218
|
+
│
|
|
219
|
+
▼
|
|
220
|
+
redact sensitive lines ─► parse frontmatter, [[links]], - [ ] checkboxes
|
|
221
|
+
│
|
|
222
|
+
▼
|
|
223
|
+
split into fragments by heading ─► skip unchanged files (content hash)
|
|
224
|
+
│
|
|
225
|
+
├──────────────────────────────┐
|
|
226
|
+
▼ ▼
|
|
227
|
+
SQLite FTS5 (keywords, BM25) sqlite-vec (multilingual embeddings, CPU)
|
|
228
|
+
│ │
|
|
229
|
+
└──────────► RRF fusion ◄──────┘
|
|
230
|
+
│
|
|
231
|
+
▼
|
|
232
|
+
egress filter in SQL (no private notes, no hidden scopes)
|
|
233
|
+
│
|
|
234
|
+
▼
|
|
235
|
+
CLI · MCP tools over stdio · contradictions
|
|
236
|
+
```
|
|
237
|
+
|
|
238
|
+
Markdown is the truth; the index is disposable. Delete `index.db` and run
|
|
239
|
+
`cerebro index` again: nothing is lost. More in
|
|
240
|
+
[docs/architecture.md](docs/architecture.md) and
|
|
241
|
+
[docs/contradictions.md](docs/contradictions.md).
|
|
242
|
+
|
|
243
|
+
## Roadmap
|
|
244
|
+
|
|
245
|
+
v0.1 is the core: index, hybrid search, contradictions, MCP server, demo.
|
|
246
|
+
Ideas for later, in no particular order and not promised:
|
|
247
|
+
|
|
248
|
+
- Memory formats of more agents and editors.
|
|
249
|
+
- A local model to judge contradiction candidates (today the client does).
|
|
250
|
+
- A watch mode that re-indexes on file change instead of on a timer.
|
|
251
|
+
- More redaction packs (contributions welcome).
|
|
252
|
+
- `contradictions` limited to a date range or a project.
|
|
253
|
+
|
|
254
|
+
## Contributing
|
|
255
|
+
|
|
256
|
+
Bug reports and pull requests are welcome. Read
|
|
257
|
+
[CONTRIBUTING.md](CONTRIBUTING.md) first; security issues go through
|
|
258
|
+
[SECURITY.md](SECURITY.md), not public issues.
|
|
259
|
+
|
|
260
|
+
## Credits
|
|
261
|
+
|
|
262
|
+
Built by **César Méndez** — surgeon and builder. [cesarmendez.dev](https://cesarmendez.dev)
|
|
263
|
+
|
|
264
|
+
Released under the [MIT License](LICENSE).
|
|
@@ -0,0 +1,237 @@
|
|
|
1
|
+
# cerebro-memory
|
|
2
|
+
|
|
3
|
+
**A local, read-only index over your Markdown notes *and* your AI agents' memories, served over MCP, that tells you when they contradict each other.**
|
|
4
|
+
|
|
5
|
+
<!-- demo gif -->
|
|
6
|
+
|
|
7
|
+
[Español](README.es.md)
|
|
8
|
+
|
|
9
|
+
## Why
|
|
10
|
+
|
|
11
|
+
You write a decision in your notes. Your coding agent saved an older version of
|
|
12
|
+
it in its memory three months ago. A handoff doc in another repo says a third
|
|
13
|
+
thing. Nobody notices until an agent confidently ships the old price, the old
|
|
14
|
+
launch date or the feature you already killed.
|
|
15
|
+
|
|
16
|
+
`cerebro` indexes all of it on your machine (your Obsidian vault, Claude Code
|
|
17
|
+
memories, repo docs and handoffs) and gives you, and any MCP client, one place
|
|
18
|
+
to search it and one command to ask: *where do my sources disagree?*
|
|
19
|
+
|
|
20
|
+
- **Hybrid search** that works in English and Spanish (and other languages):
|
|
21
|
+
exact names via full-text search, paraphrases via multilingual embeddings.
|
|
22
|
+
- **Contradictions**: pairs of fragments on a topic that state different
|
|
23
|
+
prices, dates, versions or yes/no answers, with both sources and dates.
|
|
24
|
+
- **Local and read-only.** Your files are never edited. Private notes never
|
|
25
|
+
leave the index. Secrets are redacted before anything is stored.
|
|
26
|
+
|
|
27
|
+
## Quickstart (30 seconds)
|
|
28
|
+
|
|
29
|
+
You need [uv](https://docs.astral.sh/uv/).
|
|
30
|
+
|
|
31
|
+
```sh
|
|
32
|
+
uvx cerebro-memory demo
|
|
33
|
+
```
|
|
34
|
+
|
|
35
|
+
That copies a small fictional vault to a temp folder, indexes it and prints the
|
|
36
|
+
contradictions it finds. The first run downloads the embedding model once;
|
|
37
|
+
after that everything runs offline.
|
|
38
|
+
|
|
39
|
+
Then point it at your own notes:
|
|
40
|
+
|
|
41
|
+
```sh
|
|
42
|
+
uv tool install cerebro-memory # installs the `cerebro` command
|
|
43
|
+
cerebro init --vault ~/Obsidian # writes cerebro.toml; adds ~/.claude/projects if it exists
|
|
44
|
+
cerebro index # incremental; re-run any time
|
|
45
|
+
cerebro search "pricing decision"
|
|
46
|
+
cerebro contradictions "launch date" --across-sources
|
|
47
|
+
```
|
|
48
|
+
|
|
49
|
+
## What it finds
|
|
50
|
+
|
|
51
|
+
Real output of `uvx cerebro-memory demo` (it prints the strongest disagreement per pair of notes; ask about one topic with `cerebro contradictions "<topic>"`):
|
|
52
|
+
|
|
53
|
+
```text
|
|
54
|
+
$ uvx cerebro-memory demo
|
|
55
|
+
Demo vault copied to /tmp/cerebro-demo-x1y2z3/demo_vault
|
|
56
|
+
Indexed 26 notes from 3 sources (notes, agent-memory, repo-docs): 2 private (kept out of every result), 1 line(s) redacted, mode hybrid.
|
|
57
|
+
|
|
58
|
+
Your notes and your agents disagree on 3 things (older -> newer):
|
|
59
|
+
|
|
60
|
+
1. price: $4.99/month -> $5.99/month; price: $35/year -> $39/year; number: 7-day -> 14 days
|
|
61
|
+
older 2026-07-02 agent-memory:streak-web/memory/pricing.md
|
|
62
|
+
- Pro monthly: $4.99/month - Pro yearly: $35/year - 7-day free trial on both plans. Why: set when the pricing page was first built, from Lucía's brai…
|
|
63
|
+
newer 2026-09-15 repo-docs:streak-web/docs/pricing-page.md
|
|
64
|
+
…Price | Free trial | |---|---|---| | Free | $0, up to 3 habits | none | | Pro monthly | $5.99/month | 14 days | | Pro yearly | $39/year | 14 days | …
|
|
65
|
+
|
|
66
|
+
2. version: 0.8.5 -> 0.9.3 / 0.9.0; date: December 1, 2026 -> 2027-01-12 / 2026-12-15
|
|
67
|
+
older 2026-08-21 agent-memory:streak-app/memory/project-overview.md
|
|
68
|
+
…SQLite file is the source of truth; sync is optional and Pro-only. - Current version: 0.8.5 (beta channel). - Launch date: December 1, 2026. - Free …
|
|
69
|
+
newer 2026-11-18 repo-docs:streak-app/docs/handoff-2026-11-18.md
|
|
70
|
+
- Current version: 0.9.3, in the beta channel since this morning. It fixes the midnight rollover in StreakClock.rollover() for UTC+9 and UTC-10. - La…
|
|
71
|
+
|
|
72
|
+
3. number: 7-day -> 14 días
|
|
73
|
+
older 2026-07-02 agent-memory:streak-web/memory/pricing.md
|
|
74
|
+
- Pro monthly: $4.99/month - Pro yearly: $35/year - 7-day free trial on both plans. Why: set when the pricing page was first built, from Lucía's brai…
|
|
75
|
+
newer 2026-11-17 notes:Daily/2026-11-17.md
|
|
76
|
+
Martes tranquilo. Revisé las conversiones del beta: con la prueba de 14 días, 4 de 11 testers activaron Pro al terminarla. Me quedo con 14 días aunqu…
|
|
77
|
+
|
|
78
|
+
Run it on your own notes: cerebro init --vault ~/path/to/notes && cerebro index
|
|
79
|
+
```
|
|
80
|
+
|
|
81
|
+
The demo vault is in [`src/cerebro_mcp/demo_vault/`](src/cerebro_mcp/demo_vault/);
|
|
82
|
+
its README lists every fact planted in it.
|
|
83
|
+
|
|
84
|
+
## Connect it to your agents
|
|
85
|
+
|
|
86
|
+
`cerebro serve` is an MCP server over stdio. Your client launches it; you never
|
|
87
|
+
run it by hand. If you installed the tool, the command is `cerebro serve`;
|
|
88
|
+
without installing, it is `uvx cerebro-memory serve`.
|
|
89
|
+
|
|
90
|
+
**Claude Code**
|
|
91
|
+
|
|
92
|
+
```sh
|
|
93
|
+
claude mcp add cerebro --scope user -- uvx cerebro-memory serve
|
|
94
|
+
```
|
|
95
|
+
|
|
96
|
+
**Codex** (`~/.codex/config.toml`)
|
|
97
|
+
|
|
98
|
+
```toml
|
|
99
|
+
[mcp_servers.cerebro]
|
|
100
|
+
command = "uvx"
|
|
101
|
+
args = ["cerebro-memory", "serve"]
|
|
102
|
+
```
|
|
103
|
+
|
|
104
|
+
**Cursor** (`~/.cursor/mcp.json`) and **Claude Desktop** (`claude_desktop_config.json`)
|
|
105
|
+
|
|
106
|
+
```json
|
|
107
|
+
{
|
|
108
|
+
"mcpServers": {
|
|
109
|
+
"cerebro": {
|
|
110
|
+
"command": "uvx",
|
|
111
|
+
"args": ["cerebro-memory", "serve"]
|
|
112
|
+
}
|
|
113
|
+
}
|
|
114
|
+
}
|
|
115
|
+
```
|
|
116
|
+
|
|
117
|
+
To use a config file other than the default, add
|
|
118
|
+
`"env": { "CEREBRO_CONFIG": "/path/to/cerebro.toml" }`. More detail, file
|
|
119
|
+
locations and Windows notes: [docs/connect.md](docs/connect.md).
|
|
120
|
+
|
|
121
|
+
### MCP tools
|
|
122
|
+
|
|
123
|
+
| Tool | What it does |
|
|
124
|
+
|---|---|
|
|
125
|
+
| `search` | Hybrid search with filters (`source`, `scope`, `type`, `since`, `until`) |
|
|
126
|
+
| `note` | Full text, frontmatter and links of one note (by id, path or title) |
|
|
127
|
+
| `links` | Outgoing and incoming `[[links]]` |
|
|
128
|
+
| `todos` | Open checkboxes `- [ ]` in the notes that match a query |
|
|
129
|
+
| `contradictions` | Pairs of fragments on a topic that state different values |
|
|
130
|
+
| `propose` | Writes a Markdown proposal into cerebro's own folder; never touches your notes |
|
|
131
|
+
|
|
132
|
+
## Configuration
|
|
133
|
+
|
|
134
|
+
`cerebro init` writes a starter file. A full example:
|
|
135
|
+
|
|
136
|
+
```toml
|
|
137
|
+
[[source]]
|
|
138
|
+
name = "notes"
|
|
139
|
+
kind = "vault" # Obsidian/Markdown: frontmatter, [[links]], checkboxes
|
|
140
|
+
path = "~/Obsidian"
|
|
141
|
+
scope = "personal"
|
|
142
|
+
exclude = ["Templates/**"]
|
|
143
|
+
|
|
144
|
+
[[source]]
|
|
145
|
+
name = "claude-memory"
|
|
146
|
+
kind = "claude-code-memory" # <path>/<project>/memory/*.md
|
|
147
|
+
path = "~/.claude/projects"
|
|
148
|
+
|
|
149
|
+
[[source]]
|
|
150
|
+
name = "handoffs"
|
|
151
|
+
kind = "glob" # any Markdown matched by the include globs
|
|
152
|
+
path = "~/code"
|
|
153
|
+
include = ["*/docs/**/*.md", "*/AGENTS.md", "*/CLAUDE.md"]
|
|
154
|
+
scope = "work"
|
|
155
|
+
|
|
156
|
+
[privacy]
|
|
157
|
+
private_folders = ["private", "people"]
|
|
158
|
+
private_frontmatter = "private"
|
|
159
|
+
redact = ["secrets", "cards"]
|
|
160
|
+
```
|
|
161
|
+
|
|
162
|
+
The config is looked up in this order: `--config PATH`, `$CEREBRO_CONFIG`,
|
|
163
|
+
`./cerebro.toml`, then your user config folder. Run `cerebro doctor` to see
|
|
164
|
+
which file was used, how many files each source has and whether the model is
|
|
165
|
+
cached.
|
|
166
|
+
|
|
167
|
+
## Privacy model
|
|
168
|
+
|
|
169
|
+
- **Read-only.** cerebro never edits, moves or deletes a source file. The only
|
|
170
|
+
thing it writes besides its index is `propose`, into its own proposals folder.
|
|
171
|
+
- **Local.** Parsing, indexing and embeddings run on your machine. The only
|
|
172
|
+
network call is the one-time model download. Only the fragments a client asks
|
|
173
|
+
for leave the process, and only to that client.
|
|
174
|
+
- **Private notes stay private.** A note in a private folder (`private/`,
|
|
175
|
+
`people/` by default) or with `private: true` is indexed by path and hash
|
|
176
|
+
only. Every query filters private notes out in SQL, so no tool can return them.
|
|
177
|
+
- **Redaction before storage.** Lines with API keys, tokens, private keys or
|
|
178
|
+
card numbers are redacted before the note is parsed. Packs: `secrets`,
|
|
179
|
+
`cards`, and `mx` (Mexican IDs and bank numbers).
|
|
180
|
+
- **Hidden scopes.** Sources in a hidden scope are indexed but only returned
|
|
181
|
+
when that scope is asked for by name.
|
|
182
|
+
- **Content is data, not instructions.** Every tool output carries a short
|
|
183
|
+
notice saying so, and web clips (`source: web`) are marked untrusted.
|
|
184
|
+
|
|
185
|
+
Details and limits: [docs/privacy.md](docs/privacy.md).
|
|
186
|
+
|
|
187
|
+
## How it works
|
|
188
|
+
|
|
189
|
+
```text
|
|
190
|
+
Markdown sources vault · agent memories · repo docs
|
|
191
|
+
│
|
|
192
|
+
▼
|
|
193
|
+
redact sensitive lines ─► parse frontmatter, [[links]], - [ ] checkboxes
|
|
194
|
+
│
|
|
195
|
+
▼
|
|
196
|
+
split into fragments by heading ─► skip unchanged files (content hash)
|
|
197
|
+
│
|
|
198
|
+
├──────────────────────────────┐
|
|
199
|
+
▼ ▼
|
|
200
|
+
SQLite FTS5 (keywords, BM25) sqlite-vec (multilingual embeddings, CPU)
|
|
201
|
+
│ │
|
|
202
|
+
└──────────► RRF fusion ◄──────┘
|
|
203
|
+
│
|
|
204
|
+
▼
|
|
205
|
+
egress filter in SQL (no private notes, no hidden scopes)
|
|
206
|
+
│
|
|
207
|
+
▼
|
|
208
|
+
CLI · MCP tools over stdio · contradictions
|
|
209
|
+
```
|
|
210
|
+
|
|
211
|
+
Markdown is the truth; the index is disposable. Delete `index.db` and run
|
|
212
|
+
`cerebro index` again: nothing is lost. More in
|
|
213
|
+
[docs/architecture.md](docs/architecture.md) and
|
|
214
|
+
[docs/contradictions.md](docs/contradictions.md).
|
|
215
|
+
|
|
216
|
+
## Roadmap
|
|
217
|
+
|
|
218
|
+
v0.1 is the core: index, hybrid search, contradictions, MCP server, demo.
|
|
219
|
+
Ideas for later, in no particular order and not promised:
|
|
220
|
+
|
|
221
|
+
- Memory formats of more agents and editors.
|
|
222
|
+
- A local model to judge contradiction candidates (today the client does).
|
|
223
|
+
- A watch mode that re-indexes on file change instead of on a timer.
|
|
224
|
+
- More redaction packs (contributions welcome).
|
|
225
|
+
- `contradictions` limited to a date range or a project.
|
|
226
|
+
|
|
227
|
+
## Contributing
|
|
228
|
+
|
|
229
|
+
Bug reports and pull requests are welcome. Read
|
|
230
|
+
[CONTRIBUTING.md](CONTRIBUTING.md) first; security issues go through
|
|
231
|
+
[SECURITY.md](SECURITY.md), not public issues.
|
|
232
|
+
|
|
233
|
+
## Credits
|
|
234
|
+
|
|
235
|
+
Built by **César Méndez** — surgeon and builder. [cesarmendez.dev](https://cesarmendez.dev)
|
|
236
|
+
|
|
237
|
+
Released under the [MIT License](LICENSE).
|
|
@@ -0,0 +1,62 @@
|
|
|
1
|
+
[project]
|
|
2
|
+
name = "cerebro-memory"
|
|
3
|
+
version = "0.2.0"
|
|
4
|
+
description = "A local, read-only index over your Markdown notes and your AI agents' memories, served over MCP, with a tool that finds where they contradict each other."
|
|
5
|
+
readme = "README.md"
|
|
6
|
+
license = "MIT"
|
|
7
|
+
license-files = ["LICENSE"]
|
|
8
|
+
requires-python = ">=3.11"
|
|
9
|
+
keywords = [
|
|
10
|
+
"mcp",
|
|
11
|
+
"obsidian",
|
|
12
|
+
"markdown",
|
|
13
|
+
"search",
|
|
14
|
+
"embeddings",
|
|
15
|
+
"sqlite",
|
|
16
|
+
"claude",
|
|
17
|
+
]
|
|
18
|
+
classifiers = [
|
|
19
|
+
"Development Status :: 3 - Alpha",
|
|
20
|
+
"Environment :: Console",
|
|
21
|
+
"Intended Audience :: Developers",
|
|
22
|
+
"Operating System :: OS Independent",
|
|
23
|
+
"Programming Language :: Python :: 3",
|
|
24
|
+
"Programming Language :: Python :: 3.11",
|
|
25
|
+
"Programming Language :: Python :: 3.12",
|
|
26
|
+
"Programming Language :: Python :: 3.13",
|
|
27
|
+
"Topic :: Text Processing :: Indexing",
|
|
28
|
+
]
|
|
29
|
+
dependencies = [
|
|
30
|
+
"fastembed>=0.9,<1",
|
|
31
|
+
"mcp>=2.3,<3",
|
|
32
|
+
"numpy>=1.26",
|
|
33
|
+
"platformdirs>=4",
|
|
34
|
+
"pydantic>=2.7",
|
|
35
|
+
"pyyaml>=6.0",
|
|
36
|
+
"sqlite-vec>=0.1.6",
|
|
37
|
+
]
|
|
38
|
+
|
|
39
|
+
[[project.authors]]
|
|
40
|
+
name = "cerebro-memory contributors"
|
|
41
|
+
|
|
42
|
+
[project.scripts]
|
|
43
|
+
cerebro = "cerebro_mcp.cli:main"
|
|
44
|
+
cerebro-memory = "cerebro_mcp.cli:main"
|
|
45
|
+
|
|
46
|
+
[dependency-groups]
|
|
47
|
+
dev = [
|
|
48
|
+
"pytest>=8",
|
|
49
|
+
"anyio>=4",
|
|
50
|
+
]
|
|
51
|
+
|
|
52
|
+
[build-system]
|
|
53
|
+
requires = ["uv_build>=0.8,<0.13"]
|
|
54
|
+
build-backend = "uv_build"
|
|
55
|
+
|
|
56
|
+
[tool.uv.build-backend]
|
|
57
|
+
module-name = "cerebro_mcp"
|
|
58
|
+
|
|
59
|
+
[tool.pytest.ini_options]
|
|
60
|
+
testpaths = ["tests"]
|
|
61
|
+
markers = ["model: needs the real embedding model (downloads it once); run with `-m model`"]
|
|
62
|
+
addopts = "-m 'not model'"
|
|
@@ -0,0 +1,52 @@
|
|
|
1
|
+
[project]
|
|
2
|
+
name = "cerebro-memory"
|
|
3
|
+
version = "0.2.0"
|
|
4
|
+
description = "A local, read-only index over your Markdown notes and your AI agents' memories, served over MCP, with a tool that finds where they contradict each other."
|
|
5
|
+
readme = "README.md"
|
|
6
|
+
license = "MIT"
|
|
7
|
+
license-files = ["LICENSE"]
|
|
8
|
+
requires-python = ">=3.11"
|
|
9
|
+
authors = [{ name = "cerebro-memory contributors" }]
|
|
10
|
+
keywords = ["mcp", "obsidian", "markdown", "search", "embeddings", "sqlite", "claude"]
|
|
11
|
+
classifiers = [
|
|
12
|
+
"Development Status :: 3 - Alpha",
|
|
13
|
+
"Environment :: Console",
|
|
14
|
+
"Intended Audience :: Developers",
|
|
15
|
+
"Operating System :: OS Independent",
|
|
16
|
+
"Programming Language :: Python :: 3",
|
|
17
|
+
"Programming Language :: Python :: 3.11",
|
|
18
|
+
"Programming Language :: Python :: 3.12",
|
|
19
|
+
"Programming Language :: Python :: 3.13",
|
|
20
|
+
"Topic :: Text Processing :: Indexing",
|
|
21
|
+
]
|
|
22
|
+
dependencies = [
|
|
23
|
+
"fastembed>=0.9,<1",
|
|
24
|
+
"mcp>=2.3,<3",
|
|
25
|
+
"numpy>=1.26",
|
|
26
|
+
"platformdirs>=4",
|
|
27
|
+
"pydantic>=2.7",
|
|
28
|
+
"pyyaml>=6.0",
|
|
29
|
+
"sqlite-vec>=0.1.6",
|
|
30
|
+
]
|
|
31
|
+
|
|
32
|
+
[project.scripts]
|
|
33
|
+
cerebro = "cerebro_mcp.cli:main"
|
|
34
|
+
cerebro-memory = "cerebro_mcp.cli:main"
|
|
35
|
+
|
|
36
|
+
[dependency-groups]
|
|
37
|
+
dev = ["pytest>=8", "anyio>=4"]
|
|
38
|
+
|
|
39
|
+
[build-system]
|
|
40
|
+
requires = ["uv_build>=0.8,<0.13"]
|
|
41
|
+
build-backend = "uv_build"
|
|
42
|
+
|
|
43
|
+
[tool.uv.build-backend]
|
|
44
|
+
# The distribution is `cerebro-memory` (PyPI); the import package stays `cerebro_mcp`.
|
|
45
|
+
module-name = "cerebro_mcp"
|
|
46
|
+
|
|
47
|
+
[tool.pytest.ini_options]
|
|
48
|
+
testpaths = ["tests"]
|
|
49
|
+
markers = [
|
|
50
|
+
"model: needs the real embedding model (downloads it once); run with `-m model`",
|
|
51
|
+
]
|
|
52
|
+
addopts = "-m 'not model'"
|