confos 0.4.0

Conference intelligence in your terminal — for humans and agents.

confos

Conference intelligence in your terminal — for humans and agents.

confos turns AI/ML conferences (NeurIPS, ICLR, ICML, COLM, …) into a local, queryable knowledge base. Ingest a venue once from OpenReview, then search papers, find the people working on a topic, see what's trending, visualize the landscape, and export agent-ready context — all locally, all scriptable, with provenance back to the source.

It's built for terminals, shell scripts, CI, and coding agents (Claude Code, Codex, or anything with shell access). Every command speaks stable JSON; after a one-time ingest every query is local, offline, and millisecond-fast — and unlike asking an LLM, every number is real and traceable, not guessed.

Run confos brief --venue neurips-2025 and the whole conference lands in one screen — a colored dashboard in your terminal:

Plain-text version (for piping / screen readers)

◆ confos brief · neurips-2025
  5,540 papers · 23,013 authors · 29 orgs · 9,826 topics
  ████████████████████████████████████████████████
  ▆ accepted 5286 (95%)   ▆ rejected 254 (5%)
  Status mix reflects publicly-visible submissions only — OpenReview hides most
  rejected papers, so this is not the venue's acceptance rate.

HOT TOPICS
most-published subtopics here
large language models    ████████████████████████████████████████  301
reinforcement learning   ███████████████████████████████████████▌  297
diffusion models         ████████████████████████                  181
reasoning                ██████████████▍                           108
interpretability         ████████████▎                              92

What it looks like

Find the people working on a topic — ranked, with a why-relevant reason and provenance:

See what's trending — colored topic bars straight in the terminal:

Surface the best-reviewed work — mean rating, spread, and review count from public reviews:

---

Install

Requires Python 3.12+.

uv tool install confos     # recommended — isolated global install
pipx install confos        # alternative isolated install
pip install confos         # plain pip
uvx confos --help          # run once, without installing

Upgrade later with uv tool upgrade confos (or pipx upgrade confos).

Quickstart

confos init                          # one-time: create the local store at ~/.confos
confos ingest neurips-2025           # pull a venue from OpenReview (network, a few minutes)
confos brief --venue neurips-2025    # the dashboard: topics, orgs, top papers, people

Not sure of a venue slug? confos venues search "ICLR 2026" or confos venues aliases. After the one-time ingest, everything is local and offline.

Commands

confos is organized into a few groups. Every command supports --json (a stable envelope for agents/scripts) and --plain (TSV); the default is a rich human view. Run confos <group> --help for flags, or confos schema <command> for the exact JSON contract.

Setup & data

Command	What it does
confos init	Create the local store (~/.confos).
confos ingest <venue> [--with-reviews] [--force] [--dry-run]	Pull a venue's full submission set from OpenReview. --with-reviews also fetches review scores; re-running does an incremental update.
confos enrich profiles --venue <v>	Fetch author profiles → affiliations, countries, homepage/Scholar/DBLP. Powers the org & country stats.
confos index rebuild · confos index status	Re-derive the index from the raw snapshots, offline · show row counts.
confos venues list | search <q> | show <slug> | add | aliases	Discover and register venues.
confos doctor	Check the environment, store, FTS5, and the OpenReview backend.

Search & explore papers

Command	What it does
confos papers search <query> [--venue --year --org --accepted-only --limit]	Full-text search over title/abstract/keywords, ranked and cited.
confos papers show <id> [--with related]	A reading-friendly card: authors, abstract, and links.
confos papers related <id>	Papers similar by title/keyword overlap.
confos papers top [--topic --venue]	Highest-rated papers (needs ingest --with-reviews).
confos papers controversial [--topic --venue]	The most divisive papers — highest rating variance.

confos papers search "long-running agents" --venue neurips-2025 --limit 20
confos papers search "tool use" --accepted-only --json | jq '.data[].title'

People

Command	What it does
confos authors find --topic <t> [--venue]	Rank the people actually working on a topic — with their papers, affiliation, and a why-relevant explanation.
confos authors search <name>	Find an author by name.
confos authors show <id>	Author detail.
confos authors papers <id> [--venue]	An author's papers.
confos authors coauthors <id>	Their collaborators, ranked by shared papers.

Organisations

Command	What it does
confos orgs top [--venue]	Top organisations by paper count.
confos orgs papers <org> [--venue]	Papers affiliated with an organisation.

Org & country data is populated by confos enrich profiles — coverage is sparse without it, and confos is always honest about that.

Stats & trends

Command	What it does
confos stats overview | topics | orgs | countries [--explain]	Honest aggregates — every stat reports its data-quality.
confos trends topic <t> --venues a,b,c	How a topic moves across venues / years.
confos trends compare <a> <b> --topic <t>	Two venues, head-to-head.

Visualize

Command	What it does
confos viz topics | orgs [--venue]	Colored terminal bar charts.
confos viz network --topic <t> [--format terminal|mermaid|html]	Co-authorship graph — view in-terminal or export Mermaid/HTML.

Brief & export

Command	What it does
confos brief --venue <v> [--topic <t>]	The one-command dashboard (pictured above): overview, hot topics, rising orgs, top papers, people-to-know.
confos export context --topic <t> [--format json|markdown]	A self-contained, cited context pack purpose-built for agents.
confos export papers | authors [--format csv|jsonl]	Bulk data dumps.
confos schema <command>	Print the stable JSON output contract for a command.

Output modes — human, agent, script

Three modes, switchable on any command:

Mode	Flag	For
Human	(default)	A rich, colored view — tables, the brief dashboard, clickable titles. Degrades to plain text on a pipe / NO_COLOR / a dumb terminal.
JSON	--json	A stable, versioned envelope (schema_version · data · provenance) — the contract agents and scripts depend on. confos schema <cmd> documents it.
Plain	--plain	Line/TSV output for grep / awk / cut.

stdout carries only the requested data; progress and warnings go to stderr, so a --json command stays parseable even mid-ingest.

For agents

confos is a first-class tool for coding agents (Claude Code, Codex, …):

• Stable JSON on every command, with a documented schema and provenance — the agent never has to hallucinate a statistic.
• confos export context --topic <t> returns one cited pack (top papers, ranked people, orgs, topic-scoped stats) — drop it straight into a prompt.
• A bundled skill (.agents/skills/confos/) teaches an agent how to drive the tool. See AGENTS.md.

How it works

A one-time ingest snapshots a venue's raw notes to JSONL (the source of truth) and derives a local SQLite + FTS5 index. Every query after that is local, offline, and deterministic, with each result traceable to its OpenReview source. OpenReview is the only source today; more adapters, semantic search, an LLM ask, and an MCP server are designed-for-later (the seams exist). See docs/ARCHITECTURE.md.

Documentation

Doc	What's in it
docs/PRODUCT.md	Goal, the wedge, who it's for, worked examples for every capability
docs/ARCHITECTURE.md	System design, diagrams, components, data model, stack
docs/CLI_CONTRACT.md	Full command tree, flags, output contract, exit codes, safety
docs/SCHEMAS.md	Stable --json output shapes (the agent/script contract)
docs/RANKING.md	How authors find ranks people, and how --topic matching works
docs/VISUAL.md	The human-output visual layer — theme, semantic color, the brief dashboard, links
AGENTS.md	How an agent should drive confos
CONTRIBUTING.md	Setup, the gate (ruff + mypy + pytest), layering rules

Contributing

Contributions welcome — see CONTRIBUTING.md for setup and the gate. The opt-in scripts/live-test.sh exercises the real OpenReview API before a release.

git clone https://github.com/RRaphaell/confos && cd confos
uv sync && uv run confos --help

License

MIT — see LICENSE.