Memory Tools

Every conversation with your AI starts from scratch. You re-explain your pricing strategy, your team structure, your product roadmap - every single time. Memory tools fix that.

See it in action: ask your AI to “run the memory demo” - a guided walkthrough on temporary data that cleans up after itself. More about demos.

The idea

Store important context once. Recall it anytime with a simple question. Your AI searches semantically - meaning it understands what you meant, not just the exact words you used.

“Remember: our ideal customer is a 10-50 person SaaS company doing $1M-$5M ARR.”

“Recall what we know about our target customer.”

That’s it. Stored and searchable. The next time you ask your AI to write outreach copy or evaluate a lead, it already knows who you’re going after.

---

Tools

Tool	What it does
memory_store	Save a specific piece of context (personal or org-wide)
memory_recall	Search your memories using natural language, with score, scope, project, and date filters
memory_update	Edit an existing memory in place without losing its history
memory_forget	Delete a memory you no longer need
memory_summarize_and_store	Distill a long conversation into a stored memory

Near-duplicate detection runs on every store: if a new memory is more than 92% similar to an existing one, FoundersOS surfaces the conflict and lets you force-store or skip rather than silently piling up duplicates.

---

Personal vs. org memories

Every memory has a scope:

• Personal - only visible to you (identified by your FOUNDERS_OS_USER_ID)
• Org - visible to everyone on your team who shares the same Supabase project

This matters when you’re working with a team. Org memories are shared context - company strategy, pricing decisions, product direction. Personal memories are your own notes and reminders.

"Remember for the org: we decided to sunset the free tier on March 1."
"Remember for me: I need to prep the investor deck by Friday."

---

How semantic search works

When you store a memory, FoundersOS converts it into a mathematical vector - a representation of its meaning, not just its keywords. When you recall, your question gets converted the same way, and we find the closest matches.

This means:

• Searching for “pricing” will find memories about “cost”, “rate”, and “what we charge.”
• Searching for “target market” will find memories about “ideal customer profile” and “who we sell to.”
• You don’t need to remember the exact words you used.

Tip: Think of it like this - traditional search is like looking for a book by its title. Semantic search is like describing what the book was about and having the librarian find it for you.

---

Embedding providers

The semantic search runs on vector embeddings. You can choose your provider:

Provider	Model	Dimensions	Cost
OpenAI (default)	text-embedding-3-small	1536	Pay-per-use, very cheap
Ollama	nomic-embed-text	768	Free, runs locally
AWS Bedrock	amazon.nova-2-multimodal-embeddings-v1:0	1024	AWS pricing

Caution: Pick your provider before setup. The vector dimension is set when you create the database table. If you start with OpenAI (1536 dimensions) and want to switch to Ollama (768), you’ll need to re-create the memory table and re-store your memories. Pick one and stick with it, at least for now.

---

What to store

Here are some ideas to get you started:

• Company strategy - “Our north star metric is weekly active teams.”
• Pricing decisions - “We offer 20% annual discount on all plans.”
• Team context - “Alex handles enterprise accounts, Sam owns product.”
• Product direction - “We’re not building a mobile app until we hit 1,000 paying teams.”
• Meeting takeaways - “Investor said to focus on retention before growth.”
• Process notes - “We always send a follow-up email within 24 hours of a demo.”

The more you teach your AI, the more useful it becomes. Think of it as building your company’s institutional knowledge - one conversation at a time.

---

Example session

You: "Remember for the org: we just closed our seed round 
      at $2M. Lead investor is Horizon Ventures. 
      We're using it for eng hiring and go-to-market."

AI: Stored. I'll remember that as an org-wide memory.

You: "Recall what we know about our funding."

AI: Here's what I found:
    - Seed round closed at $2M, led by Horizon Ventures.
    - Funds allocated to engineering hiring and 
      go-to-market efforts.

You: "Write a brief update email to the team 
      about our funding news."

AI: [Writes the email using the stored context, 
     no re-explanation needed]