trove-ingest

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

trove-ingest

Evidence first, meaning second. The graph never contains a claim it cannot prove.

先存证据，后赋含义。图谱中绝不会包含无法验证的断言。

When to use

适用场景

User shares a URL, file, gist, or paste worth remembering.
A long document (spec, transcript, article) should become queryable memory.

When not to use:

Conversation-born knowledge with no external source →
```
trove-remember
```
.
The source is a vault page — the importer handles those (
```
npm run import:vault
```
).

用户分享值得留存的URL、文件、gist或粘贴文本。
需要将长文档（规格说明、转录稿、文章）转化为可查询的记忆内容。

Process

不适用场景

Step 1 — obtain the text

—

Fetch/read the source. For URLs, extract the readable text.

无外部来源的对话生成知识 → 使用
```
trove-remember
```
。
来源为vault页面 —— 此类内容由导入器处理（
```
npm run import:vault
```
）。

Step 2 — store the evidence

流程

—

步骤1 — 获取文本

ingest { kind, title, uri, contentText, metadata }

```
kind
```
:
```
url
```
|
```
file
```
|
```
paste
```
|
```
transcript
```
|
```
email
```
|
```
slack
```
.
Identical content dedupes by hash — re-ingesting is safe.
Returns the
```
source
```
and its addressable
```
textUnits
```
.

获取/读取来源内容。对于URL，提取可阅读文本。

Step 3 — extract atoms

步骤2 — 存储证据

From the text units, remember the meaning-bearing atoms (typically far fewer than the text is long): entities, claims, decisions, patterns. For each:

remember { title, type, summary, evidence: [{ textUnitId }], links }

Every atom cites the exact text unit that supports it. Link atoms to existing project/domain nodes via

links

; remember's response lists similar nodes if the target already exists.

ingest { kind, title, uri, contentText, metadata }

```
kind
```
:
```
url
```
|
```
file
```
|
```
paste
```
|
```
transcript
```
|
```
email
```
|
```
slack
```
。
相同内容会通过哈希去重 —— 重复导入是安全的。
返回
```
source
```
及其可寻址的
```
textUnits
```
。

Step 4 — annotate (optional)

步骤3 — 提取原子

annotate

to mark important spans ("contradicts claim X", "important quote") without rewriting anything.

从文本单元中提取承载含义的原子（数量通常远少于文本长度）：实体、断言、决策、模式。针对每个原子执行：

remember { title, type, summary, evidence: [{ textUnitId }], links }

每个原子都会引用支持它的精确文本单元。通过

links

将原子链接到已有的项目/领域节点；

remember

的响应会列出相似节点（如果目标节点已存在）。

Step 5 — confirm

步骤4 — 标注（可选）

Source id + how many text units; atoms remembered with slugs; anything that contradicted existing nodes (recommend

trove-remember

trove-lint

使用

annotate

标记重要片段（如“与断言X矛盾”、“重要引用”），无需修改任何内容。

Anti-patterns

步骤5 — 确认

Don't remember atoms without text-unit citations when the source is right there.
Don't turn the whole document into one giant node — the source already stores the full text; atoms are the compression.
Don't ingest secrets or credentials as content.

返回来源ID及文本单元数量；返回带slug的已记录原子；返回所有与现有节点矛盾的内容（建议使用

trove-remember

或

trove-lint

处理）。

—

反模式

—

切勿在来源可用的情况下，记录不带文本单元引用的原子。
切勿将整个文档转化为一个巨型节点 —— 来源已存储完整文本；原子是对内容的压缩。
切勿将机密或凭据作为内容导入。