Historia Collect

Historia Collect is the local ingress and retrieval application for AI conversation history. It imports official ChatGPT account exports, captures the conversation currently rendered in the browser, stores both raw and normalized records as reachable Git objects, and builds a disposable SQLite search index.

Git remains authoritative. Search indexes, context bundles, summaries, and the application interface are projections that can be deleted and rebuilt.

Complete backfill

Import an official ChatGPT ZIP, extracted export, conversations.json, or numbered conversation files. Re-importing the same archive is idempotent.

Live browser collection

Capture the currently rendered ChatGPT conversation through a restricted native-messaging bridge without reading cookies, tokens, or browser storage.

Agent context

Search current or historical message revisions and construct bounded, provenance-rich context packages for Codex, Kimi, or another CLI agent.

Quick start

Install the package.
Terminal window
```
npm install -g @greenways-ai/historian
```
Initialize the private local vault.
Terminal window
```
historia vault init
```

Install the browser bridge.

historia-collect install --browser chrome

Start the local application.
Terminal window
```
historia collect serve
```
Import an official export or load the reported unpacked extension directory in the browser.

What gets stored

Each collection source advances an independent ref:

refs/historia/sources/<provider>/<source-key-digest>

A Git commit represents one atomic collection transaction—not one message. Its tree makes normalized message revisions, raw provider records, conversation graphs, source metadata, assets, and import receipts reachable.

sources/
conversations/
messages/
raw/messages/
raw/conversations/
raw/exports/
imports/

Messages retain stable logical identities while immutable Git blobs represent content revisions. Edited messages and regenerated branches therefore remain part of the historical record rather than silently replacing earlier content.

Search and context

Build or refresh the SQLite projection:

historia chat index

Search the current message revisions:

historia chat search "signed rooms"

Include older revisions:

historia chat search "signed rooms" --historical

Construct a bounded Markdown context package:

historia context build \
  "Hestia keys and private rooms" \
  --budget 12000 \
  --include-branches

Every result carries its source ref, archive commit, conversation manifest path, logical message ID, immutable revision OID, and message path.

Trust classifications

Historia distinguishes how content entered the archive:

Classification	Meaning
`full-account-export`	Provider-generated account snapshot.
`conversation-export`	Export intended to cover one conversation.
`browser-observed`	Content rendered or visited in a browser.
`cli-native`	Content emitted directly by a local CLI adapter.
`user-supplied`	Manually supplied material.
`derived`	Summaries, tags, projections, or other generated records.

A browser capture records what Historia observed. It does not claim provider authorship or account completeness.