napsack 0.1.3

NAPsack records and aggregates your computer use — screenshots plus input events (click, keypress, scroll, cursor move). It groups activity into event bursts and uses a VLM pipeline to generate human-readable captions describing what happened.

NAPsack

NAPsack records and structures your computer use by generating natural language caption from screenshots and input events (click, keypress, scroll, cursor move).

---

Quickstart

Requires Python 3.11+

Install NAPsack from PyPI:

pip install napsack

Or, if you prefer not to install it, use uv to run the module commands shown below.

API Keys

NAPsack uses a VLM to generate captions. Create a .env file in the project root (or export variables in your shell):

cp .env.example .env

NAPsack uses litellm, so pass a full provider-prefixed model string via --model:

Provider	--model example	API key variable
Gemini (default)	gemini/gemini-3-flash-preview	GEMINI_API_KEY
OpenAI	openai/gpt-4.1-mini	OPENAI_API_KEY
Anthropic	anthropic/claude-4.6-sonnet	ANTHROPIC_API_KEY
vLLM (self-hosted)	hosted_vllm/Qwen3-VL-8B + --api-base http://host/v1	(none required)
Ollama (local)	ollama/qwen3.5:4b + --api-base http://localhost:11434 (replace with yours)	(none required)
Tinfoil (confidential inference)	(pass --client tinfoil)	TINFOIL_API_KEY
BigQuery	(pass --client bigquery)	Application Default Credentials — gcloud auth application-default login

Record a session (press CTRL+C to stop)

napsack-record --session-dir ./logs/session_name --monitor
# or without installing:
uv run -m napsack.record --session-dir ./logs/session_name --monitor

Label the recorded session

napsack-label --session-dir ./logs/session_name # default model is Gemini 3.0 flash
# or without instaling
uv run -m napsack.label --session-dir ./logs/session_name 

NAPsack uses litellm as its labeling backend, supporting Gemini, vLLM/self-hosted models, and any OpenAI-compatible API (OpenAI, Together, Groq, etc.). It also supports Tinfoil for confidential inference (your data never leaves a verified TEE), and integrates with BigQuery (if you're on Screenomics, this is for you :D).

To use Tinfoil, pass --client tinfoil --model <model>, e.g.:

napsack-label --session-dir ./logs/session_name --client tinfoil --model kimi-k2-5

Output

./logs/session_name
├── screenshots         # Recorded screenshots
├── aggregations.jsonl  # Recorded event bursts
├── captions.jsonl	    # All VLM-generated captions
├── annotated.mp4       # Final video showing generated captions and input events
└── data.jsonl          # Final data containing raw input events and LLM generated captions

Method

Record

NAPsack groups temporally adjacent input events of the same type into event bursts. An event is assigned to the current burst if the time since the preceding event of that type does not exceed the corresponding gap threshold and the elapsed time since the burst start remains within the max duration.

• If the gap threshold is exceeded, a new burst is started.
• If the max duration is exceeded, the first half of the current burst is finalized and saved, while the second half becomes the active burst. A burst is force-restarted when the active monitor changes.

Label

The label module:

• Loads sessions or raw screenshots and chunks.
• Uses prompts (in label/prompts) to instruct the VLM to generate captions that describe the user's actions and context.
• Produces captions.jsonl and data.jsonl (captions aligned to screenshots and events).
• Optionally renders an annotated video (annotated.mp4) showing captions and event visualizations overlayed on frames.

The label step performs a second layer of aggregation: it uses the bursts detected at recording time and further refines and annotates them with VLM outputs to create final human-readable summaries.

Citation

If you use NAPsack in your research, please cite:

@misc{shaikh2026learningactionpredictorshumancomputer,
      title={Learning Next Action Predictors from Human-Computer Interaction},
      author={Omar Shaikh and Valentin Teutschbein and Kanishk Gandhi and Yikun Chi and Nick Haber and Thomas Robinson and Nilam Ram and Byron Reeves and Sherry Yang and Michael S. Bernstein and Diyi Yang},
      year={2026},
      eprint={2603.05923},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2603.05923},
}