Local-first collection
Known public or authorized pages can be fetched, normalized, and indexed before a paid search or extraction provider is considered.
Open source
Built by Raintree Technology
DocPull Web Evidence for AI Agents
Explore how DocPull collects local-first web evidence, checks source dependencies, enforces budget boundaries, and writes inspectable context artifacts.
Install: pip install docpull
Definition
What DocPull does
DocPull is an open-source context dependency engine for AI agents. It collects web evidence, checks whether sources are fresh and stable, preserves citations and provenance, and writes artifacts that a person can inspect outside the model conversation.
Audience
Who it is for
DocPull is for engineers and agent builders who need reproducible web context, local-first retrieval, explicit paid-provider boundaries, and a durable record of which sources supported a downstream answer.
Capabilities
Inspectable formats
Outputs include Markdown, NDJSON, SQLite, manifests, indexes, archives, and portable knowledge bundles.
Dependency checks
Source freshness, citation presence, stability, and reproducibility remain part of the evidence workflow.
How the work stays reviewable
Budget boundary
A zero-dollar budget blocks paid-capable routes before execution instead of merely reporting spend afterward.
Provider escalation
Search, extraction, and browser routes are added only when local and open discovery cannot satisfy the task.
Evidence trail
URLs, route steps, timestamps, hashes, and output paths stay with the collected material for later review.