Info Hub

One private dashboard for everything coming at you — your chat and messaging accounts, turned into tidy info units (tasks, events, topics) you can actually read.

_{Quick tour: Today for the work that needs attention, Explore for the tagged info-unit graph, and Chat for answers grounded in your organized messages. Synthetic demo data.}

Chat groups are noisy. Info Hub reads your own Lark/Feishu accounts (and any source plugin you add) locally and read-only, groups the chatter into main-idea info units, and shows you what matters on one dashboard — todos, events, a calendar, and topic summaries — with everything staying on your machine.

Screenshots: Today, Explore, Graph and Chat

_{Today — priorities, events and flagged messages. Synthetic demo data.}

_{Explore — filter units by person, topic, channel or type. Synthetic demo data.}

_{Graph — inspect connected units and open any node. Synthetic demo data.}

_{Chat — streamed answers with links to units, groups and people. Synthetic demo data.}

Try it with the demo data: python scripts/seed_demo.py writes a fictional example database to data/demo.db; point hub.database at it in config/hub.local.yaml and run python -m infohub serve.

Explore's List and Graph share one sidebar for sorting and filtering, including keywords, kinds, time, status and tags. Message totals include every subtopic branch, and graph node size follows those totals. Graph filtering is strict by default; linked context is an explicit option. See the Explore guide.

Chat can organize the graph

Ask the assistant to connect topics, add a unit under several parents, remove a connection, consolidate duplicate tags, change a unit's kind or dates, and attach message evidence. It can inspect graph edges and tag counts before editing, and reports the changes it applies. For example: “Link the reward-design topic under world models”, or “Consolidate topic:rl-training into topic:reinforcement-learning”. Chat needs an OpenAI-compatible model configured in Settings.

How it works

flowchart TB
  subgraph L1["1 - Sync (local, read-only, no AI)"]
    direction LR
    SRC["Your accounts<br/>Lark/Feishu x N orgs · pluggable sources"]
    PRE["preprocess<br/>flags · dates · segments · triage P0-P3"]
    SRC --> PRE
  end
  subgraph L2["2 - Organize into info units"]
    direction LR
    RULE["rule organizer<br/>always on, deterministic"]
    AI["AI organizer<br/>optional · budget · retries"]
    AGENT["agent bridge<br/>Codex / Claude Code"]
  end
  subgraph L3["3 - Apps"]
    direction LR
    DASH["Dashboard<br/>Today · Todos · Calendar · Groups · Explore"]
    REM["Reminders"]
    APIX["Agent API + brief"]
  end
  PRE --> DB[("SQLite + FTS<br/>messages")]
  DB --> RULE & AI & AGENT
  RULE & AI & AGENT --> UG[("info-unit graph<br/>tags · links · evidence")]
  UG --> DASH & REM & APIX
  DB -. "search, flagged lane" .-> DASH

Loading

It runs every day without AI. Layer 1 and the rule organizer are deterministic. The AI organizer is a separate, optional worker: if the API is down, over budget or not configured, its jobs wait in a queue and nothing else is affected. Full design in docs/architecture.md.

What it never does

• It sends, posts or edits nothing on any platform. Dashboard actions only change local state.
• It never commits data. data/, config/*.local.yaml and logs/ are git-ignored.
• It sends no message text to an AI service unless ai.enabled: true, and then only for chats whose policy is listed in ai.policies (default: chats you marked Watch).
• It listens only on 127.0.0.1 by default, rejects foreign Host headers and cross-origin writes, and requires login once an account exists.

Get started

With a coding agent (Claude Code, Codex, …) — clone the repo, open it in your agent and paste:

Please set up Info Hub for my Lark/Feishu organizations on this machine.
Follow docs/setup.md: run the non-interactive steps yourself, and for each browser
approval open the page and tell me what to click. I have N organizations: <names>.

By hand — follow docs/setup.md. The short version (Git, Python 3.10+, Node 18+):

git clone https://github.com/MasterYip/InfoHub.git
cd InfoHub
python -m venv .venv && . .venv/bin/activate      # Windows: .\.venv\Scripts\activate
pip install -r requirements.txt && npm install    # npm installs the official lark-cli
cp config/hub.example.yaml config/hub.local.yaml
# once per Lark/Feishu organization, approve in the browser inside that organization:
npx lark-cli config init --new --name org1
npx lark-cli auth login --profile org1 --domain im,calendar,task,contact
python -m infohub doctor
python -m infohub run

Then open http://127.0.0.1:8765/ and set your chats and names in Settings. Full steps, multiple organizations, AI, autostart and remote access are in docs/setup.md.

Windows desktop and tray

For a dedicated window, install the optional desktop packages and launch:

.\.venv\Scripts\python.exe -m pip install -r requirements-desktop.txt
.\scripts\start-desktop.ps1

Close the window to keep Info Hub in the tray. Click the tray icon to reopen it, or use Open in browser for the original web interface. The tray groups new-unit notifications, can limit them to For me, and opens the latest unit when a notification is clicked. Use scripts/install-autostart.ps1 -Desktop to start in the tray at logon. See desktop setup and behavior for lifecycle, login and notification settings.

Automatic database backups

Info Hub makes a verified database copy on first service startup, then every 7 days. Change the interval in Settings → Database backups, run a copy immediately with Back up now, and check the last successful backup and next due time there. Completed copies are kept in data/backups/ and excluded from Git. See backup configuration and recovery.

Scheduled agents (Codex, Claude Code, ...)

An external agent can do the Layer 2 work on a schedule instead of, or alongside, the API worker:

• python -m infohub agent next leases jobs as JSON (instructions, schema, jobs); agent submit results.json applies the results. Leases expire, so an abandoned run never loses work.
• python -m infohub agent context --days 1 returns a brief plus counts for other agent-team tasks. HTTP equivalents: /api/agent/next, /api/agent/submit, /api/agent/context.
• agents/scheduled-run.md is the instruction file. scripts/agent-run.ps1 (or .sh) runs agent.command from your config, e.g. codex exec --full-auto. scripts/install-agent-schedule.ps1 -Hours 2 registers a Task Scheduler entry; on Linux use the cron line in agent-run.sh.
• Set ai.queue_for_agents: true to queue work for agents while the API worker is off.

Commands

Layer	Command	Purpose
1	python -m infohub sync [--source <name>]	One sync round, plus the rule organizer (prints counts only)
2	python -m infohub organize [--ai] [--requeue DAYS]	Rule organizer; --ai runs queued AI jobs; --requeue queues past chats
2	python -m infohub agent next / agent submit FILE / agent context	Scheduled or external agents (lease protocol)
3	python -m infohub brief --days 7 --tags course:ai-engineering	Markdown brief for agent-team tasks
3	python -m infohub remind	Toast reminders for due todos and upcoming events
—	python -m infohub run	The service: all layers on their own schedules, plus the dashboard
—	python -m infohub stats / doctor	Counts and health (no message text)

The HTTP API is documented at the top of infohub/apps/server.py.

Remote access and login

To open the dashboard from other devices, create an account (python -m infohub user add <name>), forward the port through your public server, and use HTTPS. Once an account exists, every page and API call requires login; the browser keeps a 30-day session cookie. See docs/remote-access.md.

Running on an always-on machine

The hub runs anywhere lark-cli does. See docs/architecture.md § Deployment; reach it from other devices via a tunnel with login (docs/remote-access.md).