corvusmem 0.1.2 → 0.1.3
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.
- package/CHANGELOG.md +9 -0
- package/README.md +41 -0
- package/docs.md +9 -0
- package/package.json +5 -1
package/CHANGELOG.md
CHANGED
|
@@ -5,6 +5,15 @@ Format based on [Keep a Changelog](https://keepachangelog.com/); versions follow
|
|
|
5
5
|
|
|
6
6
|
## [Unreleased]
|
|
7
7
|
|
|
8
|
+
## [0.1.3] — 2026-11-04
|
|
9
|
+
### Added
|
|
10
|
+
- README: badges (npm, node, zero-deps, offline, license), a **Why Corvusmem** list, a **Usage** section with the
|
|
11
|
+
everyday commands, and Website/Docs/Benchmarks/npm links.
|
|
12
|
+
### Changed
|
|
13
|
+
- Docs (README + `docs.md`): clarified that **per-project vs global only decides where the hooks turn on**, not where
|
|
14
|
+
memory lives (project facts always land in the repo's `.corvusmem/`), plus a **team-sharing recipe** (commit
|
|
15
|
+
`.corvusmem/` and, for a per-project install, `.claude/settings.json`; never commit `~/.config/corvusmem/`).
|
|
16
|
+
|
|
8
17
|
## [0.1.2] — 2026-11-03
|
|
9
18
|
### Fixed
|
|
10
19
|
- `homepage` points to the live site `https://corvusmem.vercel.app` (the `.com` domain isn't registered yet).
|
package/README.md
CHANGED
|
@@ -4,13 +4,32 @@
|
|
|
4
4
|
|
|
5
5
|
# Corvusmem
|
|
6
6
|
|
|
7
|
+
<p align="center">
|
|
8
|
+
<a href="https://www.npmjs.com/package/corvusmem"><img src="https://img.shields.io/npm/v/corvusmem?color=e07a33" alt="npm version" /></a>
|
|
9
|
+
<img src="https://img.shields.io/node/v/corvusmem" alt="node" />
|
|
10
|
+
<img src="https://img.shields.io/badge/dependencies-0-brightgreen" alt="zero dependencies" />
|
|
11
|
+
<img src="https://img.shields.io/badge/offline-yes-brightgreen" alt="offline" />
|
|
12
|
+
<img src="https://img.shields.io/badge/license-PolyForm%20NC-blue" alt="license" />
|
|
13
|
+
</p>
|
|
14
|
+
|
|
7
15
|
**Deterministic, offline, zero-dependency memory for AI coding agents.** Corvusmem stores the stable facts your
|
|
8
16
|
agent learns, replaces and dates them, and recalls them the same way every time — in any language, with no network
|
|
9
17
|
and no servers. It plugs into where you already work: Claude Code / Codex (hooks) and a standalone CLI.
|
|
10
18
|
|
|
19
|
+
🌐 [Website](https://corvusmem.vercel.app) · 📖 [Docs](https://corvusmem.vercel.app/docs) · 📊 [Benchmarks](https://corvusmem.vercel.app/benchmarks) · 📦 [npm](https://www.npmjs.com/package/corvusmem)
|
|
20
|
+
|
|
11
21
|
> **Narrative docs (what it is, the message, the Cortex, FAQ):** [`docs.md`](./docs.md). This README is the quick
|
|
12
22
|
> reference for structure and commands.
|
|
13
23
|
|
|
24
|
+
## Why Corvusmem
|
|
25
|
+
- **Deterministic** — same input, same result. No embeddings, no vector database, no model to run recall.
|
|
26
|
+
- **Offline & private** — the memory is a plain file you own. No server, no cloud, no API key, no telemetry.
|
|
27
|
+
- **Zero-dependency** — pure Node, identical on Windows, macOS and Linux.
|
|
28
|
+
- **Any language** — keys and values in any language; recall is language-agnostic.
|
|
29
|
+
- **Per-author** — on a shared project, teammates' personal facts never overwrite each other.
|
|
30
|
+
- **Bounded footprint** — a hard cap keeps the context cost flat (O(1), not O(n)) — see the [benchmarks](https://corvusmem.vercel.app/benchmarks).
|
|
31
|
+
- **See it** — the Cortex draws your whole memory as a local HTML brain.
|
|
32
|
+
|
|
14
33
|
## How it works
|
|
15
34
|
Your agent forgets everything between sessions. Corvusmem is a memory it keeps: facts live in a lightweight local
|
|
16
35
|
graph, retrieved by name, date and relevance with two-hop activation — no embeddings, no vector database, no API key.
|
|
@@ -34,10 +53,32 @@ npx corvusmem install --dry-run # show what it would do, without writing
|
|
|
34
53
|
Merge-safe (it never overwrites your hooks or permissions) and idempotent. It creates `.corvusmem/` (project
|
|
35
54
|
memory) and `~/.config/corvusmem/` (global memory), and writes `.corvusmem/hooks.json` (a portable manifest).
|
|
36
55
|
|
|
56
|
+
**Per-project vs global only decides *where the hooks turn on*** — for one repo (`--claude`, config in that repo) or
|
|
57
|
+
for every project (`--global`, in your home). It does **not** change where memory lives: project facts always land in
|
|
58
|
+
that repo's `.corvusmem/`, your personal facts in `~/.config/corvusmem/`. You never need a per-project install just to
|
|
59
|
+
keep memory in the repo — that happens either way.
|
|
60
|
+
|
|
61
|
+
**Sharing with a team**: commit `.corvusmem/` (the shared project memory) and, for a per-project install,
|
|
62
|
+
`.claude/settings.json` so teammates get the hooks the moment they clone. Never commit `~/.config/corvusmem/` — that is
|
|
63
|
+
your private memory.
|
|
64
|
+
|
|
37
65
|
**Codex**: the hooks run through the plugin path (a marketplace in `~/.codex/config.toml`), not the project
|
|
38
66
|
`settings.json`. Publish it to a Codex marketplace, install it and approve the hooks in the app. On Claude Code they
|
|
39
67
|
are active right away.
|
|
40
68
|
|
|
69
|
+
## Usage
|
|
70
|
+
Once installed, the agent captures and recalls memory on its own — you don't do a thing. To look at it or steer it:
|
|
71
|
+
```
|
|
72
|
+
npx corvusmem --cortex # your whole memory as a local HTML brain (offline)
|
|
73
|
+
npx corvusmem --recall "stripe webhook" # ask what it already knows, from the terminal
|
|
74
|
+
npx corvusmem --list # list every fact
|
|
75
|
+
npx corvusmem --save --dimension project.stack --value "Astro + Tailwind" --scope project
|
|
76
|
+
npx corvusmem --forget <id> # drop one fact
|
|
77
|
+
```
|
|
78
|
+
The Cortex also takes `--client` (a shareable view with nothing private), `--output <dir>` and `--lang en|es|…`.
|
|
79
|
+
Flags are English with Spanish aliases (`--save`/`--guardar`, `--recall`/`--recordar`); keys and values work in any
|
|
80
|
+
language — the memory is language-agnostic.
|
|
81
|
+
|
|
41
82
|
## Test
|
|
42
83
|
```
|
|
43
84
|
npm test
|
package/docs.md
CHANGED
|
@@ -29,6 +29,11 @@ npx corvusmem install --dry-run # show what it would do, without writing
|
|
|
29
29
|
Merge-safe (it never overwrites your hooks or permissions) and idempotent. It creates `.corvusmem/` (project memory)
|
|
30
30
|
and, if requested, `~/.config/corvusmem/` (global user memory).
|
|
31
31
|
|
|
32
|
+
**Per-project or global only decides *where the hooks turn on*** — in one repo (`--claude`, the config lives in that
|
|
33
|
+
repo) or in every project (`--global`, in your home). It does **not** change where memory lives: a project's facts
|
|
34
|
+
always land in that repo's `.corvusmem/`, and your personal facts in `~/.config/corvusmem/`. You don't need a
|
|
35
|
+
per-project install just to keep memory in the repo — it goes there either way.
|
|
36
|
+
|
|
32
37
|
**Codex**: the user installs it from the Codex app (via plugin), not from the CLI. The agent does not install on Codex.
|
|
33
38
|
|
|
34
39
|
## How it works
|
|
@@ -59,6 +64,10 @@ prefers Vim and Bob prefers Emacs, both are kept and neither overwrites the othe
|
|
|
59
64
|
itself** (stack, decisions) stay shared and have no owner. You configure nothing — the author is read from your `git`
|
|
60
65
|
email (or your system user), with no network. For a solo developer it's invisible; it just works.
|
|
61
66
|
|
|
67
|
+
**To share it with the team**, commit `.corvusmem/` (the shared project memory) and, if you installed per-project,
|
|
68
|
+
`.claude/settings.json` so teammates get the hooks the moment they clone. Never commit `~/.config/corvusmem/` — that is
|
|
69
|
+
your private, personal memory.
|
|
70
|
+
|
|
62
71
|
## The Cortex
|
|
63
72
|
|
|
64
73
|
The Cortex draws everything your agent knows as a **brain of points** — one glow per fact. Each kind of memory settles
|
package/package.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "corvusmem",
|
|
3
|
-
"version": "0.1.
|
|
3
|
+
"version": "0.1.3",
|
|
4
4
|
"description": "Deterministic, offline, zero-dependency memory for AI coding agents (Claude Code, Codex) and a standalone CLI.",
|
|
5
5
|
"type": "module",
|
|
6
6
|
"bin": {
|
|
@@ -18,6 +18,7 @@
|
|
|
18
18
|
"scripts": {
|
|
19
19
|
"test": "node tests/test-boundary.mjs && node tests/test-facts.mjs && node tests/test-hard-recall.mjs && node tests/test-install.mjs && node tests/test-plugin.mjs && node tests/test-parity.mjs && node tests/test-cortex-visual.mjs",
|
|
20
20
|
"test:core": "node tests/test-boundary.mjs && node tests/test-facts.mjs && node tests/test-hard-recall.mjs && node tests/test-install.mjs && node tests/test-plugin.mjs && node tests/test-parity.mjs",
|
|
21
|
+
"bench:tokens": "node tools/eval-tokens.mjs",
|
|
21
22
|
"preversion": "npm run test:core",
|
|
22
23
|
"prepublishOnly": "npm run test:core"
|
|
23
24
|
},
|
|
@@ -45,5 +46,8 @@
|
|
|
45
46
|
},
|
|
46
47
|
"bugs": {
|
|
47
48
|
"url": "https://github.com/petersonsenadevs/corvusmem/issues"
|
|
49
|
+
},
|
|
50
|
+
"devDependencies": {
|
|
51
|
+
"gpt-tokenizer": "^4.1.0"
|
|
48
52
|
}
|
|
49
53
|
}
|