Building Todoist Ramble: How Doist Turned Voice Braindumps Into Real-Time Task Capture

Key Takeaways

• Ramble processes raw audio directly with Gemini live‑audio model, skipping transcription
• Tool‑only architecture adds, edits, deletes tasks while user is still speaking
• Full project/label list injected into prompt, avoiding complex RAG pipelines
• Multilingual eval uses 20+ recordings from 100+ employees across 35 countries
• Future plans add image capture, Apple Watch support, and automation hooks

Pulse Analysis

The launch of Ramble marks a pivotal shift in how productivity software handles unstructured input. By feeding live audio into Google’s Gemini model, Doist eliminates the traditional transcription step, allowing tasks to appear on screen the moment a user mentions them. This real‑time pipeline reduces cognitive load, turning the common "brain dump" habit into an instant, organized list, and showcases a lean AI implementation that other SaaS products can replicate without massive infrastructure overhead.

Doist’s technical choices reflect a focus on simplicity and reliability. Instead of building a retrieval‑augmented generation (RAG) system for project and label matching, the team injects the full context list directly into the model prompt, achieving accurate task categorization with fewer moving parts. The engineers also tackled the notoriously tricky problem of date handling by normalizing dates to English and fine‑tuning model temperature, ensuring the AI captures temporal intent without over‑interpreting. Their multilingual evaluation framework, built from recordings across 35 countries, guarantees consistent performance for a global user base.

Looking ahead, Ramble’s roadmap hints at broader multimodal capabilities, such as extracting tasks from images and integrating with wearables like the Apple Watch. These extensions could further blur the line between capture and execution, positioning Todoist as a central hub for AI‑augmented workflow automation. For enterprises, the feature offers a scalable way to digitize informal task creation, potentially improving project tracking, reducing manual entry errors, and freeing up employee time for higher‑value activities.