cs-grep

Fast, local search across computer science research literature.

cs-grep builds a local SQLite/FTS5 index from DBLP and gives you a clean CLI and TUI for searching papers with an expressive query language across title, authors, abstract, venue, year, rank, tag, and DOI.

Why

• Search across configurable computer science venue catalogs.
• Keep the corpus local and query it quickly with SQLite/FTS5.
• Search in the CLI, TUI, or browser, and export CSV, JSON, or BibTeX for scripts.

Install

Requires Rust 1.95 or newer.

cargo install --git https://github.com/philippnormann/cs-grep cs-grep

Or from a local checkout:

cargo install --path crates/cs-grep

Cargo installs to ~/.cargo/bin on macOS/Linux and %USERPROFILE%\.cargo\bin on Windows. Make sure that directory is on PATH.

Or use nix:

nix run "github:philippnormann/cs-grep" -- <arguments> # run once
nix shell "github:philippnormann/cs-grep"              # add to PATH

Use

cs-grep init
cs-grep update --since 2018
cs-grep --tui

In the TUI, use Tab to cycle sort modes, arrow keys to move, and Enter to open the selected paper URL.

Sort CLI results with --sort relevance, --sort year, --sort venue, or --sort rank.

Search from the shell:

cs-grep 'title:fuzzing WHERE venue:ndss AND year:2020-'
cs-grep '("side channel" OR cache) WHERE venue:CCS OR venue:SP'
cs-grep '* WHERE doi:10.1145' --fields venue,year,title,doi

More examples:

# Recent malware-detection papers in A/A* venues
cs-grep 'malware detection WHERE year:2022- AND (rank:A OR rank:A*)' --sort year

# Export matching papers as BibTeX
cs-grep 'kernel fuzz* WHERE venue:USENIX-SEC' --format bibtex > papers.bib

# Script-friendly JSON output
cs-grep 'large language model WHERE year:2023-' --format json

# Limit output for quick triage
cs-grep '(ransomware OR botnet) WHERE year:2020-' --limit 20

# Search a custom database path
cs-grep --db ./papers.db 'symbolic execution WHERE venue:ccs'

Query Language

Queries have a full-text expression followed by optional metadata filters:

cs-grep '(malware OR botnet) WHERE year:2020- AND NOT venue:CCS'

• Both sides support AND, OR, NOT, parentheses, and implicit AND.
• Text supports phrases, trailing prefixes such as fuzz*, and the fields title:, author:, and abstract:.
• WHERE supports venue:, year:, rank:, tag:, and doi:.
• * matches all papers when only metadata filters are needed.
• Year values can be 2020, 2018-2024, 2020-, or -2019.

Text NOT requires a positive text term; metadata filters can be negated directly.

cs-grep '* WHERE tag:ml AND tag:security'
cs-grep '* WHERE tag:ml OR tag:security'

Bundles and Tags

Bundles choose which venue catalogs update and enrich process. Tags filter papers by venue family during search. Bundled venues use the broad tags security, ml, nlp, cv, se, and hci; custom venues can define additional tags. The default bundles are security, ml, and se; hci is available separately.

cs-grep update --bundle security,ml
cs-grep update --bundle hci
cs-grep enrich --bundle ml

# ML venue papers
cs-grep 'mechanistic interpretability WHERE tag:ml AND year:2023-'

# Papers from ML-security venues
cs-grep '(prompt injection OR jailbreak) WHERE tag:ml AND tag:security'

Venues

The bundled venue catalogs are listed in crates/cs-grep-core/venues/.

After cs-grep init, you can extend or override the catalog with a user config.yaml:

• macOS: ~/Library/Application Support/cs-grep/config.yaml
• Linux: ~/.config/cs-grep/config.yaml
• Windows: %APPDATA%\cs-grep\config.yaml

You can also pass a specific file with --config path/to/config.yaml.

bundles selects bundled venue sets for subfields. User venues are always merged after selected bundles by id: reuse an existing id to override a bundled venue, or add a new id to extend the catalog.

bundles: [security, ml, se]

defaults:
  min_year: 2018

venues:
  - id: DIMVA
    name: Conference on Detection of Intrusions and Malware & Vulnerability Assessment
    dblp_stream: conf/dimva
    aliases: [dimva]
    rank: B
    tags: [systems, network, malware]

Then ingest and search it:

cs-grep update --venue DIMVA
cs-grep 'malware WHERE venue:DIMVA'

dblp_stream is the DBLP stream id used by the RDF endpoint, such as conf/dimva.

Abstracts

Abstract enrichment is optional, cached, and best-effort. cs-grep uses paper DOIs and DBLP paper URLs to find abstracts.

cs-grep enrich --jobs 8

No API keys are required, but keys can improve rate limits and coverage.

Variable	Used for	Get a key
OPENALEX_API_KEY	OpenAlex lookups	openalex.org/settings/api
SEMANTIC_SCHOLAR_S2_KEY	Semantic Scholar lookups	semanticscholar.org/product/api
OPENREVIEW_USERNAME / OPENREVIEW_PASSWORD	OpenReview lookups	OpenReview API docs

Set them in the shell:

export OPENALEX_API_KEY=...
export SEMANTIC_SCHOLAR_S2_KEY=...
export OPENREVIEW_USERNAME=...
export OPENREVIEW_PASSWORD=...

Or place them in a local .env file:

OPENALEX_API_KEY=
SEMANTIC_SCHOLAR_S2_KEY=
OPENREVIEW_USERNAME=
OPENREVIEW_PASSWORD=

.env is loaded automatically when present.

Web UI

cs-grep-web serves a search page for your cs-grep database. It's useful for hosting one instance for your group, so everyone can search the same database without installing anything.

cargo install --git https://github.com/philippnormann/cs-grep cs-grep-web
cs-grep-web  # http://localhost:5002

With Nix, run nix run "github:philippnormann/cs-grep#csGrepWeb". Pass --host 0.0.0.0 to allow remote access, and see cs-grep-web --help for other options. Restart the server after changing venues.

Docker

docker compose build
docker compose run --rm web cs-grep --db /data/papers.db --config /data/config.yaml init
docker compose run --rm web cs-grep --db /data/papers.db --config /data/config.yaml update --since 2018
docker compose run --rm web cs-grep --db /data/papers.db --config /data/config.yaml enrich  # optional, adds abstracts
docker compose up -d  # http://localhost:5002

The database and config live in a Docker volume. With Podman, use podman compose.

To serve a database you already have (Linux paths shown):

docker run --rm --init --user "$(id -u):$(id -g)" -p 127.0.0.1:5002:5002 \
  -v ~/.local/share/cs-grep:/data -v ~/.config/cs-grep:/config:ro cs-grep:latest \
  cs-grep-web --host 0.0.0.0 --db /data/papers.db --config /config/config.yaml

To share it with your group, put it behind a reverse proxy with TLS and authentication, and run update (and optionally enrich) from cron. After changing venues, copy the config in with docker compose cp config.yaml web:/data/config.yaml, then run update and docker compose restart.

HTTP API

GET /api/search

Parameter	Description
q	Query in the query language. Empty matches all papers.
sort	year (default), relevance, venue, or rank
limit	Results per page, 1-500 (default 120)
offset	Results to skip (default 0)

curl 'http://localhost:5002/api/search?q=fuzzing+WHERE+venue:NDSS+AND+year:2024&limit=1'

{
  "papers": [
    {
      "dblp_key": "https://dblp.org/rec/conf/ndss/BorrelloFDBQG24",
      "venue": "NDSS",
      "year": 2024,
      "title": "Predictive Context-sensitive Fuzzing.",
      "authors": "Pietro Borrello, Andrea Fioraldi, Daniele Cono D'Elia, ...",
      "doi": null,
      "url": "https://www.ndss-symposium.org/ndss-paper/predictive-context-sensitive-fuzzing/",
      "abstract": "Coverage-guided fuzzers expose bugs by progressively mutating testcases ..."
    }
  ],
  "next_offset": 1,
  "ranks": { "NDSS": "A*", "CCS": "A*" }
}

For the next page, pass next_offset as offset. It is null on the last page. ranks maps each venue to its rank.

GET /api/bibtex?key=<dblp_key>

Returns the paper's BibTeX entry as plain text.

Errors

Errors come back as JSON with a 4xx or 5xx status:

{ "error": "limit must be between 1 and 500" }

Acknowledgements

Inspired by top4grep.

License

Released under the MIT License.