roxabi-sense
Local workstation attention journal that tracks focus, idle, and agent sessions, exposing timeline data via MCP for AI agents to query current or past activity.
README
roxabi-sense
Workstation attention journal — light sensors, local store, agent surfaces.
Status: scaffold · design locked · implementation not started
Not screenpipe. Not a web SaaS. Not insideroxabi-factory.
Why
Screen capture + OCR is the wrong tool for “what was I doing?”.
You already have timestamped work in ~/.claude and ~/.grok. Meetings live in Claap. What is missing is a cheap focus spine: which app/window was active, which agent sessions were open, whether Slack/Discord was running — without pixels, keyloggers, or a 174 MB trial for 40 seconds of frames.
roxabi-sense is that spine. It publishes facts, not policy.
What it is / is not
| Is | Is not |
|---|---|
Local user-session daemon (systemd --user) |
Factory hub module |
| CLI + optional MCP + optional NATS publisher | Screen OCR / continuous screenshots |
| Reads existing Claude/Grok session artifacts | Re-logs AI conversations |
| Focus / idle / process presence | Meeting recorder (→ Claap) |
| Edge sensor for Sentinelle later | Sentinelle decision brain (→ factory hub) |
Architecture (target)
collectors (facts only)
┌─────────────┐ ┌──────────────┐ ┌────────────────┐
│ focus/idle │ │ agent sessions│ │ process presence│
│ (Wayland / │ │ ~/.claude │ │ slack/discord │
│ AT-SPI) │ │ ~/.grok │ │ (running?) │
└──────┬──────┘ └──────┬───────┘ └───────┬────────┘
│ │ │
└────────────────┼───────────────────┘
▼
local store (SQLite)
~/.local/share/roxabi-sense/
│
┌───────────────┼────────────────┐
▼ ▼ ▼
CLI MCP NATS (opt-in)
day-slice agent query factory.event.host.*
status what_was_i… activity | stale only
Boundary (ADR-091-aligned): sensors publish facts; roxabi-factory Sentinelle (hub module, not shipped yet) may consume host events and decide. This repo never opens Discord, never dispatches jobs, never applies ops policy.
| Surface | Granularity | Depends on factory? |
|---|---|---|
| Local store + CLI | Fine timeline | No |
| MCP (stdio) | Query on demand | No |
| NATS plane ① | Coarse heartbeats (activity / stale) |
Yes (bus up) |
Details: docs/ARCHITECTURE.md · purpose: docs/PURPOSE.md
Stack decision
Local daemon + CLI + MCP — not a Silex web boilerplate.
| Candidate | Verdict |
|---|---|
silex-boilerplate / Next / multi-page web app |
No — that shape is for client demos & product UI, not a user-session sensor |
Full stack inside roxabi-factory |
No — factory owns Sentinelle consumer; capture stays on the workstation |
Python 3.13 + uv + systemd --user |
Yes — matches Roxabi satellites (voiceCLI, xcli, cortex packages) |
SQLite under ~/.local/share/roxabi-sense/ |
Yes — single-machine, Syncthing-friendly if paused |
| MCP stdio | Yes — Claude / Grok / Hermes |
| NATS publisher (optional extra) | Yes later — facts only, feature-flagged |
Tiny 127.0.0.1 status page |
Optional V2 — human glance only, not the product surface |
Setup model (target)
# 1. install
git clone git@github.com:Roxabi/roxabi-sense.git
cd roxabi-sense && uv sync
# 2. enable user service (writes unit under ~/.config/systemd/user/)
uv run sense install-service
systemctl --user enable --now roxabi-sense.service
# 3. human CLI
uv run sense status
uv run sense day --from 14:00 --to 16:00
# 4. agents (Claude Code / Grok) — MCP stdio
# mcpServers.sense.command = ["uv", "run", "--directory", ".../roxabi-sense", "sense", "mcp"]
# 5. optional NATS (when factory Sentinelle is ready)
# sense config set nats.enabled true
# sense config set nats.url nats://...
No Podman required on the laptop for V1. M₂ may use the same user unit. M₁ host-sensor path (services snapshot only) is a later collector, not a Quadlet of this whole app.
Roadmap (coarse)
| Phase | Deliverable |
|---|---|
| 0 — this scaffold | Public repo, purpose, architecture, empty package |
| 1 — local spine | Agent-session collector + store + CLI status / day |
| 2 — focus | Wayland/COSMIC focus + idle (AT-SPI first) |
| 3 — MCP | 3–5 tools (active_now, what_was_i_doing, agent_sessions, …) |
| 4 — NATS opt-in | factory.event.host.{machine}.activity|stale for Sentinelle |
| 5 — optional | Process presence, filtered browser history, local status HTTP |
Out of scope forever (for this repo): OCR, continuous screenshots, keylogging, clipboard dumps, scraping Slack/Discord desktop clients, meeting transcription.
Relation to the rest of Roxabi
| Project | Relation |
|---|---|
roxabi-factory |
Future consumer (Sentinelle hub module). Not the home of collectors. ADR-091 factory-host-sensor role lives here as an edge process. |
roxabi-cortex |
Downstream memory/insight may ingest sense observations later. Sense stays capture + query, not the entity graph. |
| Claap | Meetings — do not duplicate |
~/.claude / ~/.grok |
Read-only sources for agent presence |
License
AGPL-3.0-or-later (same family as roxabi-cortex).
Status
Scaffold only — no collectors yet. Design intent is frozen enough to implement phase 1 without another product loop.
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.