bhr — browser history reading

日本語

Turn your Chrome history into a per-day list of pages you actually read. Work tools, search pages, SNS timelines, localhost and the like are filtered out by a config file, and repeated visits to the same page collapse into one line.

Works with Chrome and Chromium-based browsers (Brave, Edge, Vivaldi, Arc) by pointing chrome_root at their profile directory.

Install

cargo install bhr
# or, from a clone
cargo install --path .

Usage

bhr                                # TUI (default)
bhr list --day 2026-09-28          # one day as Markdown (`list` is an alias of `report`)
bhr list --since 2026-09-01        # a range (--until also works)
bhr list --json --since 2026-09-01 # [{"date", "links": [{"host", "title", "url"}]}]
bhr list --since 2026-09-01 --ask "Rust memory management"  # rank each day's pages with Jev
bhr searches --day 2026-09-28      # what you searched that day, and the pages opened from the results
bhr searches --json                # [{"date", "searches": [{"term", "links": [{"host", "title", "url"}]}]}]
bhr hosts --top 50                 # hosts by count — find what to deny
bhr hosts --json                   # [{"host", "count"}]
bhr config                         # print the config path

--config, --since, --until and --json work with every subcommand.

Searching from the CLI

Text search needs nothing extra — each page is one line in Markdown, and JSON works with jq:

bhr list --since 2026-09-01 | grep -i rust
bhr list --json --since 2026-09-01 | jq '[.[] | .links |= map(select(.title | test("rust"; "i")))]'

For meaning-based search use --ask (see Semantic search). Each day is sorted by relevance, and every JSON link gets a score (0–1). Pages scoring below min_score in the config (default 0.7) are dropped, in both --ask and the TUI's ?.

Piping bhr list into the jev CLI does not work for this: jev treats all of stdin as one state and returns one answer, not a score per page.

Search terms

searches reads what you typed into the search box (Chrome's keyword_search_terms), once per term per day, in the order you first searched. Under each term are the pages opened from its results page (one hop along from_visit, so a page reached through a redirect can be missed). The terms themselves are not filtered, but those pages go through the same filters as list, so clicks to denied hosts (Amazon, say) and top pages do not show.

The TUI shows the selected month's or day's terms in a pane under the pages, with the hosts opened from each (up to 8 lines, no scrolling).

TUI keys

Key	Action
j / k, ↓ / ↑	move
h / l, Tab	switch between days and pages
/	filter by text (all words must appear in title, host or URL; Esc clears). ↑/↓/Tab still move while typing
?	semantic search with jev: ranks the current day's pages by relevance
Enter	open the page in the browser (xdg-open / open)
x	add the selected page's host to deny (written to the config)
q / Esc	quit

Semantic search

? in the TUI (selected day) and list --ask (every day in range) work the same way. Each page (title, host, URL) is sent to Jev with a yes/no question — "is this what someone searching for query wants?" — and the list is sorted by that probability. Combine with / to narrow the pages first.

One request per page, 8 in parallel; the TUI waits until all answers are back. With --ask, narrow the range with --since / --day — a month of history can be thousands of requests. The provider follows the jev CLI: JEV_PROVIDER = typesafe (default, TYPESAFE_API_KEY), cloudflare (CLOUDFLARE_ACCOUNT_ID, CLOUDFLARE_API_TOKEN) or vercel (AI_GATEWAY_API_KEY).

Note: this sends those page titles and URLs to the provider.

Config

~/.config/bhr/config.toml (respects $XDG_CONFIG_HOME). Created with defaults on first run.

chrome_root = ""     # empty = OS default (~/.config/google-chrome, ~/Library/Application Support/Google/Chrome)
profiles = []        # empty = Default and every "Profile N"
allow = []           # empty = everything not denied; otherwise only these hosts
deny = ["github.com", "x.com", "google.com"]   # hosts to drop, subdomains included
tracking_params = ["utm_*", "fbclid", "ref"]   # query params to strip; trailing * = prefix

Always dropped regardless of config: localhost, IP addresses, .local / .test / .internal hosts, auth pages (auth., login., accounts. …), top pages (/), and iframe visits.

Tip: deny-only mode lets work domains through. If you only want reading material, list the sites you read in allow.

How it works

• The History SQLite file is copied to a temp dir before opening (Chrome locks it while running).
• Dates are local time.
• Within a day, pages are deduplicated by host + title, keeping the shortest URL.
• Chrome drops history older than ~90 days, so export regularly if you want to keep it.