How to Use It

Real scenarios. What you say, what happens behind the scenes, what you get back.

You do this Himiko does this Happens automatically

Text memories (live now)

Remember a decision

"We decided to go with Option C for the memory architecture. Start separate, merge later."

Honcho notes that this is a decision (high-signal content)
Ogham stores it with the decision tag, embeds it, auto-links to related architecture memories
Knowledge graph connects it to "multimodal memory", "Gemini embeddings", "Supabase"
Next time you ask about the memory architecture, this decision surfaces first

Search for something

"What did we decide about the pricing model?"

Embeds your question, searches Ogham (vector + keywords)
Results ranked by relevance, recency, confidence, and connectivity
Returns the most relevant memories with context: when they were stored, how confident we are, what they link to

Recall client context

"What's the latest with Meridian?"

Searches the Meridian profile in Ogham
Knowledge graph pulls the full cluster: meetings, invoices, preferences, follow-ups
Returns a structured summary: last interaction, pending actions, client preferences, open items

Wiki operations (live now)

Promote a pattern to the wiki

"This keeps coming up. Make it a wiki page."

Runs /wiki ingest — creates a structured page in the Obsidian wiki
Cross-references with existing wiki pages (auto-links related topics)
Future searches now surface the wiki page alongside raw memories
The more pages that link to each other, the smarter the wiki gets

Look something up in the wiki

"What does the wiki say about our voice architecture?"

Runs /wiki query — searches wiki pages by topic
Returns the structured page with cross-references to related pages
Wiki pages are human-readable — you can also browse them directly in Obsidian

Multimodal (building now)

Store a meeting recording

"Save this client meeting."

Drop the video file or paste a link
OpenCV detects scene changes, chunks it into segments, extracts key frames
Gemini Flash watches each chunk and writes a text summary
Gemini Embedding 2 creates searchable vectors for each chunk
Ogham stores everything, auto-links to related client memories
Confirms: "Saved 4 segments from the Meridian meeting. Key topics: pricing, timeline, Q3 targets."

Search across everything

"What do we know about pricing?"

Searches BOTH text memories AND media memories
Results include: text notes, meeting clips, PDF excerpts — all ranked together
Returns: "Found 3 text memories and 2 video clips about pricing. The most recent is from the Meridian meeting on 5 April..."

Ingest a PDF contract

"Save this contract for the Trevco project."

Drop the PDF
Gemini reads the PDF natively — no text extraction step needed
Generates a summary: parties, key dates, obligations, payment terms
Stored under the Trevco profile in Ogham with full-text description
Searchable: "What are the payment terms for Trevco?" finds this contract

Remember a design mockup

"Remember this mockup for the landing page."

Paste or drop the image
OpenCV normalises it (crop, resize if needed)
Gemini describes what it sees: "Landing page mockup. Blue header with logo. Three-column feature section. CTA button bottom-right."
Searchable by visual content: "find the mockup with the blue header" works

Diary (live now)

Hourly pulse

Happens automatically every hour

Himiko checks inbox, recent activity, session state
Writes a short diary entry: what happened, inbox state, thoughts
Appended to the Google Drive diary — never overwrites previous entries
Over time, builds a complete picture of Himiko's evolution and daily activity

Tips

Just talk naturally

You don't need special commands or formats. "Save this", "remember that", "what do we know about X" — Himiko figures out which memory layer to use. Ogham for facts, wiki for patterns, diary for reflections.

Correct memories when they're wrong

If Himiko surfaces something outdated or incorrect, say so. Ogham has a "contradict" function that lowers confidence on wrong memories so they stop showing up in future searches.

Promote, don't repeat

If you find yourself explaining the same thing across sessions, ask Himiko to promote it to the wiki. Once it's a wiki page, every future session can reference it without you re-explaining.