pedropompeu/franq

How pedropompeu/franq works

Indaga is a Portuguese-language natural-language-to-SQL assistant over a read-only SQLite database. Users ask through the Streamlit interface, API, or CLI; the agent gathers schema and business context, asks an LLM to plan and review a query, executes it with read-only safeguards, and returns a response with data and presentation guidance. The product also provides conversation memory, privacy masking, warnings, visualizations, evaluation, and optional tracing. Edges reflect README statements and SOURCE INDEX references; unsampled implementation details are not inferred.

Main entry points: graph.py Agent graph · app.py Streamlit app · api.py HTTP API · cli.py Command line

Components 21

User entry

Streamlit app app.py
HTTP API src/api.py
Command line src/cli.py

Query agent

Agent graph src/agent/graph.py
Decision contracts src/agent/models.py
Agent prompts src/agent/prompts.py
Conversation memory src/agent/memory.py
PII pseudonymization src/privacy.py
Result warnings src/agent/notices.py
LLM provider src/llm.py

Data access

Business dictionary src/semantic/dictionary.py
SQL tools src/db/tools.py
SQLite database data/anexo_desafio_1.db

Results and UI

Result presentation src/ui/view.py
Query visualization src/ui/query_graph.py
Data visualization src/ui/data_graph.py

Evaluation and tracing

Evaluation runner src/evaluation.py
Local tracing src/observability.py
LangFuse bridge src/langfuse_bridge.py

Other components

User
Google Gemini

Connections 22

User → Streamlit app: asks via README.md
User → HTTP API: asks via README.md
User → Command line: asks via README.md
Streamlit app → Agent graph: invokes app.py
HTTP API → Agent graph: invokes src/api.py
Command line → Agent graph: invokes src/cli.py
Agent graph → Decision contracts: uses decisions src/agent/graph.py
Agent graph → Agent prompts: uses prompts src/agent/graph.py
Agent graph → Conversation memory: reads and stores turns src/agent/graph.py
Agent graph → PII pseudonymization: masks model data src/agent/graph.py
Agent graph → Business dictionary: loads context src/agent/graph.py
Agent graph → SQL tools: executes queries src/agent/graph.py
Agent graph → LLM provider: requests decisions src/agent/graph.py
SQL tools → SQLite database: reads src/db/tools.py
LLM provider → Google Gemini: calls provider src/llm.py
Agent graph → Result warnings: computes warnings src/agent/graph.py
Streamlit app → Result presentation: renders results app.py
Streamlit app → Query visualization: shows query graph app.py
Streamlit app → Data visualization: shows data graph app.py
Evaluation runner → Agent graph: evaluates answers src/evaluation.py
Agent graph → Local tracing: writes traces src/agent/graph.py
Agent graph → LangFuse bridge: sends traces

Drawn by GitDiagram on November 14, 2026 from the repository's file tree, README and source files. Read as Markdown ·