How vivek1106-04/agentdb works
agentdb provides engine-aware analytics guidance through adapters, query and plan analysis, context, memory, and an MCP tool catalog. Its companion agenteval package benchmarks natural-language-to-SQL systems across ClickHouse and Databricks: suites supply questions, systems produce attempts, engines execute queries, and scoring and reporting turn traces into results. The README documents these product capabilities; some subsystem internals are unsampled, so uncertain implementation links are omitted.
Main entry points: cli.py Benchmark CLI · tools Analytics Tools · advisor Query Advisor · base.py Adapter Contract
Components 24
Agent Interface
- Tool Catalog
src/agentdb/server/app.py - Analytics Tools
src/agentdb/server/tools - Server Transport
src/agentdb/server/transports.py - Configuration
src/agentdb/config.py
Engine Access
- Adapter Contract
src/agentdb/adapters/base.py - ClickHouse Adapter
src/agentdb/adapters/clickhouse.py - Databricks Adapter
src/agentdb/adapters/databricks.py - Databricks Client
src/agentdb/adapters/databricks_client.py - Analytics Warehouses
Analytics Guidance
- Context Builder
src/agentdb/core/context_builder.py - Grounding Context
src/agentdb/core/context.py - Query Shape
src/agentdb/core/query_shape.py - Plan Analysis
src/agentdb/core/plan_analyzer.py - Query Advisor
src/agentdb/core/advisor
Query Memory
- Exemplar Store
src/agentdb/core/memory/store.py
Benchmark Harness
- Benchmark CLI
eval/agenteval/cli.py - Task Suites
eval/agenteval/tasks.py - Agent Systems
eval/agenteval/systems - Model Clients
eval/agenteval/models - Query Execution
eval/agenteval/execution.py - Benchmark Engines
eval/agenteval/engines - Blind Scoring
eval/agenteval/scorer.py - Traces and Reports
eval/agenteval/report.py
Other components
- Agent Client
Connections 25
- Agent Client → Tool Catalog: calls tools
- Tool Catalog → Analytics Tools: dispatches
src/agentdb/server/app.py - Server Transport → Tool Catalog: serves
- Analytics Tools → Context Builder: requests context
- Analytics Tools → Query Advisor: requests advice
src/agentdb/server/tools/advice.py - Analytics Tools → Exemplar Store: uses memory
- Context Builder → Grounding Context: builds
src/agentdb/core/context_builder.py - Context Builder → Adapter Contract: reads engine data
src/agentdb/core/context_builder.py - Adapter Contract → ClickHouse Adapter: implemented by
src/agentdb/adapters/clickhouse.py - Adapter Contract → Databricks Adapter: implemented by
src/agentdb/adapters/databricks.py - Databricks Client → Databricks Adapter: supports client
src/agentdb/adapters/databricks_client.py - ClickHouse Adapter → Analytics Warehouses: queries
- Databricks Client → Analytics Warehouses: executes statements
src/agentdb/adapters/databricks_client.py - Query Advisor → Query Shape: analyzes SQL
src/agentdb/core/advisor/clickhouse_advisor.py - Query Advisor → Plan Analysis: analyzes plans
- Query Advisor → Adapter Contract: uses engine capabilities
src/agentdb/core/advisor/clickhouse_advisor.py - Benchmark CLI → Task Suites: loads tasks
eval/agenteval/cli.py - Benchmark CLI → Agent Systems: runs systems
eval/agenteval/cli.py - Agent Systems → Model Clients: uses models
eval/agenteval/models/anthropic.py - Agent Systems → Query Execution: executes attempts
eval/agenteval/execution.py - Query Execution → Benchmark Engines: executes queries
eval/agenteval/engines/connect.py - Benchmark Engines → Analytics Warehouses: queries
- Benchmark CLI → Blind Scoring: scores attempts
- Benchmark CLI → Traces and Reports: generates reports
eval/agenteval/cli.py - Blind Scoring → Traces and Reports: supplies scores