Source code scan for ucbepic/docetl: 1 call site, 0 unique models in code ().
Payloop scanned the source tree of ucbepic/docetl for AI call sites — direct provider SDK use, framework wrappers, and tool-use definitions. 1 call site were found across 0 distinct models.
Cost-relevant signals: prompt caching not in use, token counting missing, streaming absent.
Source-scan grade C-.
Source Code Analysis
Source code analysis is needs improvement
How we scored this
Found 1 API call site across the codebase. No cost optimization signals found.
Adding token counting could improve cost efficiency.
LLM Locations
Models in Code
Providers
Cost Signals
Token Counting
tiktoken, count_tokens, or num_tokens usage
Prompt Caching
Anthropic cache_control or prompt caching
Batch API
OpenAI Batch API usage
Response Caching
GPTCache, LLM cache, or similar
Streaming
Streaming responses (stream=true)
Structured Output
JSON mode or response_format usage
Vision Input
Image/vision API inputs detected
| File | Provider | Pattern | Model | max_tokens | Stream | Hot Path |
|---|---|---|---|---|---|---|
| docetl/operations/utils/api.py | OpenAI | ChatCompletion | — | — | — | hot |
| Model | Input $/1M tokens | Output $/1M tokens | Note |
|---|---|---|---|
| gpt-4 | $30.00 | $60.00 | Consider cheaper alternative |
| gpt-4o | $2.50 | $10.00 | |
| gpt-4o-mini | $0.15 | $0.60 | Cost-efficient |
Highlighted models are premium-tier. For simple tasks (classification, extraction, summarization), consider routing to cheaper models like gpt-4o-mini ($0.60/1M output) or claude-3.5-haiku ($1/1M output).
