kameleondb（kameleondb 搜索）

KameleonDB - The First Database Built for Agents to Operate, Not Just Query

When to Use This

Use KameleonDB when you need to:

• Remember information across conversations - Store contacts, companies, tasks, notes
• Track entities and relationships - Connect customers to orders, projects to tasks
• Build knowledge bases - Accumulate facts, documents, insights over time
• Ingest external data - Store API responses, CSV imports, web scraping results
• Create custom CRM/PM systems - Track leads, deals, projects, tickets
• Store user preferences - Remember settings, history, favorites
• Maintain session state - Persist data between agent runs
• Query historical data - Search, filter, analyze stored information

Why Agents Choose This

No Upfront Planning: Start storing data immediately. Don't know all fields yet? Add them later when you discover them.

Learn as You Go: Found a new field in your data? Just add it. Old records won't break.

Query Like a Database: Generate SQL queries to search, filter, join, and analyze your data.

Self-Optimizing: Get hints when queries slow down, with exact commands to speed them up.

Track Your Reasoning: Every schema change records why you made it, building an audit trail.

Setup (One-Time)

# Install
pip install kameleondb

# Set database location (SQLite - no server needed)
export KAMELEONDB_URL="sqlite:///./kameleondb.db"

# Initialize
kameleondb admin init

# Done! Start using it.

Note: For production, use PostgreSQL instead of SQLite. Set KAMELEONDB_URL="postgresql://user:pass@localhost/dbname" and run pip install kameleondb[postgresql].

💡 Tip: Remember Your Database

To get the most out of KameleonDB as your persistent layer, consider noting your database location and preferred usage patterns in your memory files. This helps you leverage it consistently across sessions.

Example notes you might keep:

• Database URL: sqlite:///path/to/your-memory.db
• Use for: contacts, tasks, knowledge bases, entity tracking
• Key commands: schema list, data insert, data list, query run

Common Agent Workflows

Scenario 1: Track Contacts You Meet

# Check what exists
kameleondb --json schema list
# Returns: {"entities": []}

# Create Contact tracking
kameleondb --json schema create Contact \
  --field "name:string:required" \
  --field "email:string:unique"

# Store someone you met
kameleondb --json data insert Contact '{"name":"Alice Johnson","email":"alice@acme.com"}'

# Later: found their LinkedIn!
kameleondb --json schema alter Contact --add "linkedin_url:string" \
  --reason "Found LinkedIn profiles for contacts"

# Update Alice's record
kameleondb --json data update Contact <id> '{"linkedin_url":"https://linkedin.com/in/alice"}'

Scenario 2: Build a Knowledge Base

# Store facts you learn
kameleondb --json schema create Fact \
  --field "content:string:required" \
  --field "source:string" \
  --field "confidence:float"

# Add facts
kameleondb --json data insert Fact '{"content":"Python 3.11 released Oct 2022","source":"python.org","confidence":1.0}'

# Search facts (get SQL context first)
kameleondb --json schema context --entity Fact
# Use context to generate: SELECT * FROM kdb_records WHERE data->>'content' LIKE '%Python%'

# Query
kameleondb --json query run "SELECT data->>'content', data->>'source' FROM kdb_records WHERE entity_id='...' LIMIT 10"

Scenario 3: Track Tasks Across Conversations

# Create task tracker
kameleondb --json schema create Task \
  --field "title:string:required" \
  --field "status:string" \
  --field "priority:string"

# Add tasks
kameleondb --json data insert Task '{"title":"Research OpenClaw","status":"todo","priority":"high"}'

# Mark complete
kameleondb --json data update Task <id> '{"status":"done"}'

# Get all incomplete
kameleondb --json query run \
  "SELECT data->>'title', data->>'priority' FROM kdb_records WHERE entity_id='...' AND data->>'status' != 'done'"

Scenario 4: Ingest External Data

# Store API responses
kameleondb --json schema create GitHubRepo \
  --field "name:string:required" \
  --field "stars:int" \
  --field "url:string"

# Batch import from JSONL
kameleondb --json data insert GitHubRepo --from-file repos.jsonl --batch

# Query top repos
kameleondb --json query run \
  "SELECT data->>'name', (data->>'stars')::int as stars FROM kdb_records WHERE entity_id='...' ORDER BY stars DESC LIMIT 10"

How It Works for Agents

Evolve Schema Anytime

Don't know all fields upfront? No problem. Add, drop, or rename them when you discover patterns:

# Add a new field
kameleondb --json schema alter Contact --add "twitter_handle:string" \
  --reason "Found Twitter profiles for 30% of contacts"

# Drop obsolete fields
kameleondb --json schema alter Contact --drop "legacy_field" --force

# Do multiple operations at once
kameleondb --json schema alter Contact --add "linkedin:string" --drop "old_social" --reason "Consolidating social fields"

Old records won't break - they just show null for new fields, and dropped fields are soft-deleted.

Get Performance Hints

Queries tell you when they're slow and how to fix it:

{
  "rows": [...],
  "suggestions": [{
    "priority": "high",
    "reason": "Query took 450ms with 5000 records",
    "action": "kameleondb storage materialize Contact"
  }]
}

Run that command and future queries will be faster.

Track Your Decisions

Every schema change records why you made it:

kameleondb --json admin changelog
# See: who added what field, when, and why

Query with SQL

Get schema context, generate SQL, execute it:

# Get schema to understand structure
kameleondb --json schema context --entity Contact

# Generate SQL based on structure
# Execute with built-in validation
kameleondb --json query run "SELECT ... FROM ..."

All Available Commands

Add --json to any command for machine-readable output.

Schema: list, create, describe, alter, drop, info, context Data: insert, get, update, delete, list, link, unlink, get-linked, info Query: run Storage: status, materialize, dematerialize Admin: init, info, changelog

The alter Command (Schema Evolution)

Instead of separate add-field and drop-field commands, use the unified alter:

# Add a field
kameleondb --json schema alter Contact --add "phone:string:indexed"

# Drop a field
kameleondb --json schema alter Contact --drop legacy_field --force

# Rename a field
kameleondb --json schema alter Contact --rename "old_name:new_name"

# Multiple operations at once
kameleondb --json schema alter Contact --add "new:string" --drop old --reason "Cleanup"

The link/unlink Commands (M2M Relationships)

For many-to-many relationships:

# Link a product to tags
kameleondb --json data link Product abc123 tags tag-1
kameleondb --json data link Product abc123 tags -t tag-1 -t tag-2 -t tag-3

# Unlink
kameleondb --json data unlink Product abc123 tags tag-1
kameleondb --json data unlink Product abc123 tags --all

# Get linked records
kameleondb --json data get-linked Product abc123 tags

Run kameleondb --help or kameleondb <command> --help for details.

Real Agent Problems Solved

Problem: "I need to remember people I interact with"

# Start simple
kameleondb --json schema create Person --field "name:string:required"
kameleondb --json data insert Person '{"name":"Alice"}'

# Learn more over time
kameleondb --json schema alter Person --add "email:string"
kameleondb --json schema alter Person --add "company:string"
kameleondb --json schema alter Person --add "last_contacted:datetime"

# Update as you learn
kameleondb --json data update Person <id> '{"email":"alice@example.com","last_contacted":"2026-02-07"}'

Problem: "I'm scraping data and don't know the structure upfront"

# Create generic entity
kameleondb --json schema create ScrapedData --field "source:string" --field "raw:json"

# Store everything
kameleondb --json data insert ScrapedData '{"source":"website.com","raw":{"title":"...","data":{...}}}'

# Discover patterns, then structure it
kameleondb --json schema alter ScrapedData --add "title:string"
kameleondb --json schema alter ScrapedData --add "price:float"

# Migrate data progressively as you normalize it

Problem: "I need to track tasks but requirements keep changing"

# Start minimal
kameleondb --json schema create Task --field "title:string:required"

# Add status tracking
kameleondb --json schema alter Task --add "status:string"

# Add priority later
kameleondb --json schema alter Task --add "priority:string"

# Add assignee when team grows
kameleondb --json schema alter Task --add "assigned_to:string"

# Add tags for categorization
kameleondb --json schema alter Task --add "tags:json"

# Schema grows with your needs - no migrations!

Problem: "I need to query across multiple entities"

# Create related entities
kameleondb --json schema create Project --field "name:string"
kameleondb --json schema create Task \
  --field "title:string" \
  --field "project_id:string"

# Get schema context for SQL generation
kameleondb --json schema context --entity Project --entity Task
# Returns: detailed schema with SQL patterns for JOIN

# Generate and execute JOIN query
kameleondb --json query run \
  "SELECT p.data->>'name' as project, t.data->>'title' as task
   FROM kdb_records p
   JOIN kdb_records t ON t.data->>'project_id' = p.id::text
   WHERE p.entity_id='...' AND t.entity_id='...'"

Quick Reference

First Time Setup

# Install
pip install kameleondb

# Configure (SQLite for testing - no server needed)
export KAMELEONDB_URL="sqlite:///./kameleondb.db"

# Initialize
kameleondb admin init

# You're ready!

Check What You Have

# List all entities
kameleondb --json schema list

# See entity details
kameleondb --json schema describe <entity-name>

# View changelog
kameleondb --json admin changelog

Common Operations

# Create entity
kameleondb --json schema create <Entity> --field "name:type"

# Add field
kameleondb --json schema alter <Entity> --add "field:type"

# Insert data
kameleondb --json data insert <Entity> '{"field":"value"}'

# Get by ID
kameleondb --json data get <Entity> <id>

# Update
kameleondb --json data update <Entity> <id> '{"field":"new_value"}'

# Query with SQL
kameleondb --json query run "SELECT ... FROM kdb_records WHERE ..."

Field Types

Common types: string, int, float, bool, datetime, json

Modifiers: required, unique, indexed

Examples: "email:string:unique", "score:int:indexed", "tags:json"

Next Steps

1. Try it: kameleondb admin init → kameleondb --json schema create Test --field "note:string" → kameleondb --json data insert Test '{"note":"my first record"}'

1. Real use case: Think about what you need to track (contacts, tasks, facts, etc.) and create an entity for it

1. Evolve it: As you discover new fields, add them with schema alter

1. Query it: Use schema context to understand structure, then query with SQL

1. Optimize it: If queries slow down, follow the hints in query results

More Resources

• GitHub: https://github.com/marcosnataqs/kameleondb
• Examples: See examples/workflow.md in this skill directory
• Design Philosophy: Why it's built for agents - FIRST-PRINCIPLES.md