How lakshit-chiranjiv/contextlab works
Context Lab is a Streamlit chat app for testing how context-management strategies shape multi-turn LLM conversations. In normal mode, the app selects a strategy, estimates the derived prompt’s tokens, calls Groq, displays turn statistics, and persists conversation data and run metrics. Compare Mode runs separate conversation threads through all three strategies. Session persistence and run logging are documented capabilities; the excerpts show the SQLite layer, but omit some normal-mode orchestration details, so those links rely on README documentation.
Main entry points: app.py Single-strategy chat · llm_service.py Groq caller · database.py SQLite persistence · summarization.py Summarization
Components 14
Chat experience
- Streamlit app
app.py - Single-strategy chat
app.py - Compare Mode
app.py - Turn statistics
app.py
Context strategies
- Sliding Window
strategies/sliding_window.py - Summarization
strategies/summarization.py - Token Budget
strategies/token_budget.py
LLM services
- Token estimation
services/token_service.py - Groq caller
services/llm_service.py
Persistence
- SQLite persistence
db/database.py - Conversation sessions
db/database.py - Run log
db/database.py
Other components
- Chat user
- Groq API
Connections 20
- Chat user → Streamlit app: sends messages
README.md - Streamlit app → Single-strategy chat: dispatches chat
README.md - Streamlit app → Compare Mode: runs strategies
app.py - Single-strategy chat → Sliding Window: applies strategy
app.py - Single-strategy chat → Summarization: applies strategy
app.py - Single-strategy chat → Token Budget: applies strategy
app.py - Compare Mode → Sliding Window: applies strategy
app.py - Compare Mode → Summarization: applies strategy
app.py - Compare Mode → Token Budget: applies strategy
app.py - Single-strategy chat → Token estimation: estimates tokens
README.md - Compare Mode → Token estimation: estimates tokens
app.py - Token Budget → Token estimation: counts tokens
strategies/token_budget.py - Single-strategy chat → Groq caller: sends prompt
README.md - Compare Mode → Groq caller: sends prompts
app.py - Summarization → Groq caller: requests summary
strategies/summarization.py - Groq caller → Groq API: calls completions
services/llm_service.py - Streamlit app → SQLite persistence: loads and saves
README.md - SQLite persistence → Conversation sessions: stores sessions
db/database.py - SQLite persistence → Run log: stores turn metrics
db/database.py - Streamlit app → Turn statistics: displays metrics
README.md