A new startup wants to help developers create customized, contextual coding assistants that can connect with any model and integrate seamlessly with their development environments.

Founded in June 2023 by CEO Ty Dunn and CTO Nate Sesti (pictured above), Y Combinator alum Continue has already garnered some 23,000 stars on GitHub and 11,000 Discord community members over the past couple of years. To build on this momentum, Continue is announcing version 1.0 of its product, supported by a fresh $3 million in seed funding.

Continue’s launch comes amid an explosion in AI coding assistants like GitHub Copilot and Google’s Gemini Code Assist, not to mention younger upstarts such as Codeium and Cursor, which have raised bucketloads of cash from investors.

Continue, for its part, pitches itself as “the leading open-source AI code assistant” that can connect with any model and lets teams add their own context by pulling in data from platforms like Jira or Confluence.

With their models and context connected, developers can create custom autocomplete and chat experiences directly inside their coding environment. Autocomplete, for instance, provides in-line code suggestions as they type, while chat allows users to ask questions about a specific piece of code. The edit function also enables users to modify code by describing what changes they want to make.

The product facet of today’s announcement includes the first “major” release of Continue’s open-source extensions for VS Code and JetBrains.

“This signals to enterprises that this is a stable project you can bet on and build on,” Dunn told TechCrunch in an interview.

Separately, Continue is also launching a new hub, which can be likened to something like Docker Hub, GitHub or Hugging Face — a place for developers to create and share custom AI code assistants, replete with a registry for defining and managing the various building blocks they’re made from.

At launch, the hub includes pre-built AI coding assistants, as well as “blocks” from verified partners Mistral and its Codestral model; Claude 3.5 Sonnet from Anthropic; and DeepSeek-R1 from Ollama. However, any individual vendor or developer can contribute blocks and assistants to the hub.

A block here could mean models, which let you specify which AI model to use and where; rules for customizing the AI assistant; context to define the external context provider (e.g. Jira or Confluence); prompts to pack pre-written model prompts for invoking complex instructions; docs to define documentation sites (e.g. Angular or React); data, which allow developers to send development data to a pre-defined destination for analytical purposes; or MCP servers, which define a standard way of building and sharing tools for language models.

The idea behind this new hub is that the majority of users won’t require deep customizations — they’ll only need to make minor tweaks to coding assistants or blocks that already exist in the hub.

This raises the question: What is the incentive for creating customizations and sharing them with the world? As it turns out, it’s exactly what drives open-source communities elsewhere. Many of the launch partners are the very companies that create the underlying tools or models (e.g. Mistral and Anthropic), making Continue’s new hub an ideal place to curry favor with developers.

Moreover, the “open source ethos” is at the heart of what Continue is striving for. So if someone has created any customizations for use at work, then why not just share it with the wider community? Ultimately, Continue is positioning itself as the antithesis of proprietary “black box” AI assistant providers.

“This is a hub for the entire ecosystem to come together and work together,” Dunn said. “Instead of everybody building their own closed-source AI code Assistant, what if we had an open architecture where all of us can work together to create the building blocks people need to build tailored experiences for themselves?”

This is what Dunn refers to as establishing a “culture of contribution,” whereby developers are encouraged to experiment and create their own customizations while generating value for everyone.

“With Continue 1.0, we are enabling this culture of contribution for developers to create and share custom AI code assistants,” Dunn said. “This registry will be a place of discovery within and across organizations, which will grow in lock-step with the evolution of blocks and open, AI-enhanced developer tools.”

Then there is the data control aspect. In a more generic “one-size-fits-all” platform, the vendor can extract significant value from observing how developers operate at scale, and feed this decision-making data back into the platform to improve things for everyone. This type of activity has created controversy for the likes of GitHub Copilot, which has been accused of hijacking the hard work of millions of open source software developers for its own gains.

With Continue, the idea is that companies have more control over what happens with their data — they can share as much or as little as they like.

“When you use Continue, you get to keep your data,” Dunn said. “As an organization, you can pool all of your data for all of your developers in one place. That is not possible in the one-size-fits-all, black box code assistant, where their SaaS offerings and strategy is to take your data and use it to improve it for everyone.”

It’s still relatively early days for Continue, but the startup says it has worked with a handful of well-known businesses through the development phase — Ionos, (also an early Continue customer), as well as Siemens and Morningstar.

While large businesses are very much in its focus, Dunn says that Continue is targeting developers of all shapes and sizes, from freelancers and small teams through the gamut of enterprises. This points to how Continue will make money — its new hub ships with a free solo tier, but organizations that need greater control over their data can pay to access additional administration, governance, and security tooling.

“There’s a lot of interest from larger organizations, but we’ve also seen everything down to the individual developer who just wants some kind of customization for themselves. In those cases, I think the solo tier will be more than sufficient,” Dunn said. “But as that freelancer or small team starts to grow, and they need some amount of governance, then they can become customers.”

The free solo tier ships with three “visibility” levels. A developer’s contributions can be kept private, shared internally as part of a team, or made entirely public. Indeed, the solo tier can technically be used in a team setup; it just lacks some of the features that a team would typically require. A separate “teams” tier adds additional “multi-player” smarts to the mix, with admin controls for governing all the blocks and assistants — who has access to what.

The enterprise tier, meanwhile, ramps the data, security, and governance options up a notch with more granular controls over what blocks, models, versions, and vendors are used.

“The admin can also manage the security around credentials, where the data goes, and receive an audit log for the who, what, when and where of developer usage,” Dunn said.

Continue had previously raised $2.1 million after graduating from Y Combinator in late 2023, and it has now raised a further $3 million in SAFEs (funding with delayed equity allocation) led by developer-focused VC firm, Heavybit.

Dunn says the bulk of the fresh cash will go toward software engineering salaries, and it plans to “at least double” its current headcount of five.

“We’re using open source as a distribution approach, and so as a result, we keep our costs very low — we don’t need to capitalize nearly as much as other competitors,” Dunn said.