Unlisted · noindex · current state

Roxy — current system & setup status

Current as of August 13, 2026

Roxy is Bob’s private research assistant on Nous Hermes Cloud. She is being built to collect Bob’s research, remember how he works, and answer questions from his own source library.

What exists today

The foundation is live. Real Substack mail reaches Roxy, and the private research library already holds a first corpus of full-text documents with daily automatic ingest.

Live

Roxy on Hermes Cloud

Bob’s separate private research assistant, hosted on Nous Hermes Cloud.

Live

Bob’s interfaces

Bob can work with Roxy through Telegram and Hermes Desktop.

Live

Honcho memory

Honcho carries Bob’s preferences and working context. Woken and confirmed active (API + deriver) ahead of the August 14 session. Hosted tenants can sleep when idle; resume at the Honcho app if it naps again. Separate from the research-document library.

Live

Research intake

Roxy has a dedicated Gmail account. Bob’s Substack-only filter forwards publisher mail to Roxy. Real publisher mail is in the agent inbox, and library ingest from that mailbox is operational.

Live

Research library

Bob owns a Supabase Pro database. Schema v1 is live with full-text search. As of August 11, the corpus holds 28 Substack full-text documents from a mailbox backfill, with duplicate protection and a daily automatic library ingest.

Live

Files and recovery

Dropbox holds Roxy’s outputs, a practical rebuild kit, and the shared Inbox for large source files Bob wants Roxy to ingest. Inbox is a handoff, not a watched folder—Bob drops a file and asks.

How the system flows

Green components are live. Mail reaches Roxy and feeds the private library. Searchable corpus is not the same as every research question answered with citations and honest misses.

Planned work remaining

Next session order: (1) prove cited answers from the live library (cited hit + honest miss); (2) build company event ingestion. Pre-flight done: Honcho is already awake for the session. Mail path and 28-doc corpus are already live—do not re-prove an empty intake path. Matt Pocock skills are done: 7 safe skills live, no further installs planned. Relationship mapping is abandoned as a planned item—smarter retrieval covers the practical need at far lower complexity; the design notes are archived if real use ever demands it.

Primary

Prove cited answers from the library

Next build

Build company event ingestion

Bob gives Roxy a shareholder meeting, earnings call, investor day, conference presentation, or management interview as captions, transcript, URL, or a file in the shared Dropbox Inbox. Chat attachments stay for small text; large video/audio goes through Dropbox because Telegram silently rejects files over 20 MB. Roxy uses supplied captions first and Bob-owned ElevenLabs Scribe v2 when transcription is required, stores the timestamped research record in Supabase, verifies search and citations, then deletes provider copies and the Inbox source file.

Detailed build plan for Roxy →

Done

Matt Pocock skills — complete at 7

7 safe Matt Pocock skills are live on production Roxy. This work is finished: no further installs are planned. The remaining skills from the set of 25 stay uninstalled; dangerous and caution-flagged skills stay blocked.

Tentative plans

These are possibilities, not commitments. The next product decision should follow Bob’s real use rather than a prebuilt phase tree.

Tentative

Trusted library

The corpus is no longer empty (28 real Substack items, dedupe working). Trust still means cited answers Bob relies on, coverage reporting, and honest misses—not bulk storage alone.

Tentative

Smarter retrieval — hybrid search & chunking

Today Roxy finds documents by matching the words in Bob’s question against the words in the library. That works when the words line up, and fails silently when they don’t: ask about “datacenter power constraints” and a document that says “grid interconnect queues” never gets found—so Roxy honestly reports a miss on something the library actually contains.

The planned fix has three small parts, all inside Bob’s existing Supabase—no new services:

Not urgent at 28 documents; worth doing before the library holds hundreds. Becomes a build decision only after cited answers prove useful to Bob.

Detailed smarter-retrieval plan →

Tentative

Weekly research digest

One concise, source-cited digest after the library proves useful to Bob.

Tentative

Client portfolio analysis

Analyze masked client portfolios for concentration, research exposure, and changes over time.

Decision for Bob: retain masked snapshots for comparisons, or delete each raw upload after analysis?
Tentative

Market-data lookup

Fetch end-of-day prices on demand with a named source and as-of date, rather than building a standing market-data warehouse.

Tentative

Forecast tracking

Record dated theses with confidence, horizon, invalidation conditions, supporting sources, and later outcome review.

Tentative

Paper portfolio experiments

Test hypothetical positions separately from Bob’s real and client portfolios.

Optional · education

Local Qwen on Bob’s Mini — LM Studio

Optional teaching lab, not a production brain and not tomorrow’s work. Install LM Studio on Bob’s M4 Mini (64 GB) with Qwen 3.6 27B (MLX 8-bit). Qwen 3.8 27B is not released yet — do not download placeholder copies. LM Studio is the easy tool, not the fastest inference engine.

Start the local server only when Bob wants it; leave it off the rest of the time. Roxy on the Mini’s Desktop app can use it at localhost. Cloud/Telegram Roxy cannot see that machine unless we later add a private tunnel. If the local model is off, Roxy just says so and keeps using her Cloud model.

Detailed local-Qwen plan →

Possible path from here

Open Roxy’s detailed Company Event Ingestion build contract →