Writrieve is a local-first personal context and grounded writing engine that brings a user’s own context into the applications where they already write.

Instead of giving an LLM unrestricted access to personal data, Writrieve uses an orchestration layer to determine what context is actually needed, retrieves it through authorized sources such as Gmail and Google Calendar via Composio, normalizes and verifies the evidence, builds a task-specific writing specification, and passes only the relevant evidence to a local open-weight Qwen3-1.7B writing model.

Writrieve also maintains a provenance-aware personal memory and writing-style profile, allowing generated text to reflect the user’s relationships, facts, preferences, and writing patterns without mixing personal data directly into the underlying model.

The generated draft then passes through verification before being inserted directly into the user’s writing environment.

Core pipeline:
Task → Orchestration → Context Retrieval → Evidence State → Writing Blueprint → Qwen3-1.7B → Verification → Draft

Writrieve is designed to work across writing environments such as Gmail, LinkedIn, Substack, and other web-based editors through a Chrome extension.