Synapse Deck · Build dossier

What the code says

• The Router carries HID over raw L2CAP sockets: PSM 17 control, PSM 19 interrupt; VID:PID 1209:EDA2

  Source src/synapse_deck/infrastructure/hid_bluetooth.py

• Router: 3 hotkeys (Ctrl+F1 USB, F2 Bluetooth, F3 pairing) and 3 Pi 400 LEDs across 8 states

  Source src/synapse_deck/application/interactor.py

• OS: exactly 10 text-transformation agents are declared in the configuration

  Source synapsedeck/config/mappings.json

• OS: 8 n8n slots (F5–F12), 7 device-switch keys, 3 layouts (en-US, ru-RU, de-DE)

  Source synapsedeck/config/mappings.json

• OS: 8 numbered top-level modules plus 22 shared modules

  Source synapsedeck/modules/

• Both implementations target a Raspberry Pi 400 on 64-bit Raspberry Pi OS

  Source README.md, .env.example

• SynapseDeck OS v28.2.2; its eight numbered modules rely on asyncio, threading and subprocess alone — no machine-learning or database library is imported

  Source synapsedeck/modules/

• The "Neural Core Engine" is predictive-typing heuristics with accept/reject suggestion counters driven by the config, not a trained network

  Source synapsedeck/modules/03_neural_core_engine.py

• "Context memory" is written to context_memory.json as phrase and pattern counters; there are no embeddings and no vector index

  Source synapsedeck/modules/shared/context_memory.py

Module state

1. implemented

   HID L2CAP transport and the BlueZ profile

   An asyncio transport, an SDP record (UUID 0x1124), a pairing agent and a HIDP command handler.

   Source src/synapse_deck/infrastructure/

2. implemented

   Input capture and layout remapping

   evdev with EVIOCGRAB, a pure-Python state machine and a remap table that includes the KEY_102ND substitution for macOS.

   Source src/synapse_deck/config/layout.yaml

3. implemented

   OS backend and web dashboard

   A FastAPI app with auth/health/history/system/setup routes and a React dashboard on port 5000 with a live feed.

   Source synapsedeck/src/synapsedeck/api/

4. partial

   USB gadget "dual connection"

   The code exists but ships with USE_USB_GADGET=false and a note that Pi 400/500 are Bluetooth-only.

   Source synapsedeck/modules/shared/usb_gadget.py

5. partial

   PGP "stealth" mode

   Payload text is encrypted while keystrokes still cross an ordinary HID link; the feature is off by default.

   Source synapsedeck/modules/07_pgp_encryption_layer.py

6. not implemented

   Vector memory, MCP sync, voice and hardware accessories

   The OS tree contains no chromadb, no MCP server and no TTS or microphone code; these are marketing-page sections.

   Source synapsedeck/

7. on roadmap

   Mouse support, an on-device MCP server and Pi 500 profiles

   VISION.md v2.0.0 lists these as phases 40–45; no matching modules exist.

   Source docs/VISION.md

8. implemented

   Predictive typing and context memory

   Module 03 completes text from frequency and patterns in context_memory.json; the "neural engine" is rules and counters, with no ML library.

   Source synapsedeck/modules/03_neural_core_engine.py

9. partial

   The page’s "AI agents" versus the configuration

   The site claims "unlimited AI agents", while config/mappings.json declares 10 text-transformation agents; the module-03 framework is predictive prompts, not autonomous LLM agents.

   Source synapsedeck/config/mappings.json

How this was checked

• Router and OS are two separate repositories and two lines of development

  Different git remotes and different main-line state; CANON.md fixes the canonical folder for each.

• Four "do not touch" transport invariants are documented

  A 10-byte PSM 19 report, a 500 ms GET_REPORT delay, Logical Maximum 26 FF 00 and silence on connect.

• Router tests require a live device

  stress_test.py, test_full_matrix.py and audit_bt.sh run over SSH on the Pi with an active Bluetooth session.

  Source tests/

• The advertised OS "E2E suite" is effectively empty

  tests/e2e holds only __init__.py; no test_*.py files were found in the OS tree.

• The page’s meta description promises "unlimited AI agents" and voice, while the OS tree has no LLM client, no TTS engine and no MCP server

  A search across modules and shared finds "MCP" and "tts" only as substrings in the comments of modules 01 and 06; no implemented classes exist.

  Source synapsedeck/modules/

Limits of these claims

• Neither implementation runs without a physical Raspberry Pi 400 with its keyboard and Bluetooth radio; there is no hosted service.
• "Zero latency" and "works on every OS" are argued from the Bluetooth HID 1.1 specification and manual macOS Sequoia checks, not cross-platform CI.
• The Router has no automated unit tests, only hardware integration runs.
• n8n automation requires an external n8n instance with configured webhooks; ready-made workflows are not delivered.
• Git lineages diverge, so feature and version parity between the Router and OS trees is not guaranteed.
• The marketing page exists as loose files outside version control, so its edits are untracked.
• The published subdomain is the marketing site, not the SynapseDeck OS runtime; no module runs in a browser, so the page shows intent and its claims are read against this dossier.

Live build

Live buildsynapse-deck.moonsugar.space ↗

Revision

1412905

Build directory

site

This subdomain serves the project’s own marketing site, pinned by a content digest rather than a commit, because the source is a loose file tree outside git. Several statements on the page (agent count, vector memory, voice, hardware accessories) are not implemented in the code; the module states in this dossier are authoritative.

The page itself does not confirm availability: the host is checked when the build is published.