Semantic Search

Semantic Search

The Problem

You logged a decision about "PostgreSQL for ACID transactions" six months ago.

Now you search for "database reliability" and can't find it. The words don't match.

Continuity's semantic search understands meaning, so you can find decisions even when you use different words.

How to Search

Ask your AI to search:

You: "Search for decisions about database reliability."

AI: "Found 3 related decisions:

1. •decision-23: PostgreSQL for ACID transactions (score: 0.82)
2. •decision-45: Database connection pooling (score: 0.65)
3. •decision-67: Daily backups to S3 (score: 0.58)"

You asked about "reliability" and found decisions about "ACID transactions" and "backups" — related concepts, different words.

Search Examples

By concept

You: "Search for decisions about making the app faster."

Finds decisions about:

• •Redis caching
• •CDN usage
• •Database optimization
• •Lazy loading

By problem

You: "What decisions relate to slow page loads?"

Finds decisions that solve the problem, even if they don't mention "slow":

• •Caching decisions
• •Query optimization
• •Asset compression

By area

You: "What security decisions do we have?"

Finds decisions tagged or related to security:

• •Password hashing approach
• •Token expiry settings
• •Rate limiting

Understanding Results

When your AI searches, it shows relevance scores:

• •0.8+ — Very strong match
• •0.5–0.8 — Related
• •0.3–0.5 — Loosely related
• •Below 0.3 — Probably not relevant

You: "Show me the details for decision-23."

AI: "Decision-23: Why PostgreSQL over MongoDB?

• •Answer: ACID transactions required for payment reliability. MongoDB's eventual consistency doesn't work for financial data.
• •Tags: database, postgresql, payments
• •Created: 3 months ago"

Tips for Better Searches

Be specific

Less effective: "Database decisions"

More effective: "What did we decide about database performance and scaling?"

Describe the problem

Less effective: "Redis decisions"

More effective: "What did we decide for handling high traffic?"

Use natural language

Less effective: "auth jwt session"

More effective: "Why did we choose JWT instead of sessions?"

How It Works

Continuity uses MiniLM embeddings — a local AI model that understands meaning.

1. •Decisions are converted to vectors (384 dimensions)
2. •Your search query becomes a vector
3. •Similar meanings have similar vectors
4. •Results ranked by similarity

All processing is 100% local — no API calls, no data sent anywhere.

When to Search

Before making decisions

You: "Before we pick a state management library, search for any related decisions."

AI: "Found decision-12 about avoiding global state. No state management library decisions yet."

When onboarding

You: "What are the most important architectural decisions in this project?"

When debugging

You: "Search for decisions about error handling."

When something seems wrong

You: "We're having auth issues. What did we decide about authentication?"

Key Takeaways

1. •Ask to search — "Search for decisions about X"
2. •Use natural language — Describe what you're looking for
3. •Check scores — Higher means more relevant
4. •Search by problem — Find solutions by the problem they solve
5. •Works locally — No external APIs, full privacy

← Decision Logging | Session Notes →