Research & Announcements
Research and news
Dispatches from the frontier: launches, benchmark results, and engineering notes from Canada's sovereign frontier lab.
DCode vs Cursor vs Copilot: why we side with the agent
Autocomplete, the AI-first editor, or the full agent: a straight comparison of the three philosophies of AI-assisted coding, written by the people who built the third one.
Read the post
npm install and go: the DCode CLI and the real cost of the DipoleML API
The same Raven agent is one npm install away, and the API behind it publishes its price to the cent: $0.50 per 1M input tokens on Flash, credits that never expire, and per-key spend limits.
Read the post
OpenAI gave up on plugins. We didn't give up on integrations.
ChatGPT plugins died in 2024, and the autopsy matters: sandboxed islands with no real account access were never going to work. Connectors, done properly, are the version that survives.
Read the post
Read my last email: how Raven connects to Gmail, Calendar, Drive, and GitHub
Type "read my last email" into Raven and it does, because connectors give the model real access to your actual accounts through real OAuth consent and tightly scoped read actions.
Read the post
One account, every screen: Raven on the web, desktop, and your phone
The same Raven, the same chats, everywhere you are: a full web app, native Windows and macOS apps, and Raven on Android and iPhone, syncing live across every screen.
Read the post
Your chats are not training data: a two-minute privacy check before you paste
At Dipole ML Inc., your conversations are never used for training and inference is zero-retention by default. Here is our privacy model in plain terms, plus a two-minute checklist for any AI app.
Read the post
Raven remembers: account-level memory you can open, edit, or delete
Raven's account-level memory carries what matters across every conversation and every device: auto-captured where useful, always visible, and yours to edit or delete at any time.
Read the post
Ask Raven for a deck. Get a real .pptx.
Ask Raven for a presentation or a report and the download is a real file: PowerPoint, Word, Excel, PDF, Markdown, or CSV, assembled on your device and ready for the tools you already use.
Read the post
Deep research with receipts: how Raven reads the web and cites its sources
Ask Raven a question that needs the live web and it asks permission, reads the pages, and comes back with an answer whose numbered sources you can click and verify yourself.
Read the post
How we priced Raven: free to start, $20 Pro, and credits that never expire
The full reasoning behind our pricing: a real free tier, Pro at $20, Ultra at $100, API credits from $5 that never expire, and annual rates locked lower for Founding Members.
Read the post
A whole team, one AI: shared chats that feel like sitting at the same desk
Invite the whole team into one Raven conversation: everyone types, everyone sees every message the moment it lands, and Raven answers in the middle of the group, live on every screen.
Read the post
Share a chat like a doc: links, permissions, and who's in the room
Every Raven chat can be shared like a document: private links that stay blurred until you approve them, public links for signed-in readers, and a presence eye that shows who is in the room.
Read the post
Introducing the Founding Member Cohort: Accelerating Autonomous Intelligence
The limited launch cohort for Founding Members is open: locked-in first-year annual pricing across Pro, Ultra, and DCode, plus a permanent golden badge.
Read the post
What is sovereign AI, and why Canada needs its own frontier lab
Sovereign AI is intelligence you can verify end to end: models and infrastructure under your own jurisdiction, with privacy enforced by architecture rather than policy promises.
Read the post
A million tokens of room to think: what big context is actually for
Both Raven models carry a one-million-token context window and 128,000 tokens of output. Here is what that actually holds, and the workflows it changes once the model stops forgetting.
Read the post
Inspectable Reasoning and Zero-Retention Privacy: Inside the Raven System
A technical deep dive into the system surrounding the Raven model family: transparent reasoning streams, zero data retention, client-side DEK encryption, and prompt injection defense.
Read the post
The Dipole ML Inc. API Gateway: Native Dual Compatibility for OpenAI and Anthropic SDKs
The DipoleML API at api.dipoleml.com speaks both OpenAI and Anthropic SDK formats natively, so migrating is a one-line base URL change with zero code rewrites.
Read the post
DCode 1.0 Generally Available: The Agentic Development Environment with Native MCP
DCode 1.0 ships for macOS, Windows, and the npm CLI: native MCP support, deep codebase graphs, self-healing shell execution, and Windows computer use.
Read the post
Raven Flash Performance Report: Sub-Second Velocity and #1 on AutomationBench
Raven Flash takes #1 on AutomationBench (49.2), leads the velocity tier on Terminal Bench 4.0 (34.1), and ties #1 on the Intelligence Index, at $0.50 per 1M input tokens.
Read the post
Evaluating Raven Max: #1 on Terminal Bench 4.0 and Agent's Last Exam
The full technical evaluation of Raven Max: #1 on Terminal Bench 4.0 (42.8), Agent's Last Exam (39.5), and GDPVal-AA (1,771), with maximum reasoning effort enabled.
Read the post
Frontier Intelligence, Canadian Values: Dipole ML Inc. Officially Launches Raven and DCode
The official launch media release: Raven Flash, Raven Max, DCode 1.0, and the API gateway are live, built and served from Canada on three pillars: access, sovereignty, and trust.
Read the post
Dipole ML Inc. Announces Upcoming October 1 Launch of Sovereign Frontier Intelligence
Founder Saurabh Sharma announces the October 1, 2026 public launch of Dipole ML Inc.: the Raven model family and DCode, after four years of research.
Read the post