The recent startup wants to help programmers in creating adapted, contextual coding assistants who can connect with any model and easily integrate with their programming environments.
Founded in June 2023 by CEO You Dunn and CTO Nate Sesti (in the photo above), combinator alun To proceed He has already won about 23,000 stars on github AND 11,000 members of the community of disagreement Over the past few years. To use this rush, the continuation declares version 1.0 of its product, supported by fresh $ 3 million seed financing.
Coding assistant explosion
The introduction of continuous appears among the explosion of AI coding assistants, similar to Github Copilot and Google’s Gemini Code Assist, not to mention younger Upstarts, similar to Codeium and the cursor that raised money from investors.
Continue, for your part, puts himself as a “leading assistant of the AI Open Source” code, which may connect to any model and allows teams to add your personal context, downloading data from platforms similar to Jira or Confluence.
By combining models and context, programmers can create non -standard impressions of auto -complete and chat directly in the coding environment. AutocompleteFor example, it provides suggestions in the code line as you write, while chat It allows users to ask questions about a specific code. . edit The function also allows users to modify the code by describing what changes they need to make.
The product of the product of today’s announcement includes the first “main” edition of Open Source extensions for the VS and Pyszni code.
“It signals for enterprises that this is a stable project that you can put up and fooled,” said Dunn in an interview with TechCrunch.
Separately, the continuation also introduces a recent one centerwhich might be compared to something like Docker Hub, Github or Hulging Face – a place for programmers to create and share non -standard AI assistants, filled with defining and managing various components.
During the premiere of the center includes a pre -built AI coding assistantsAnd also “blocks” from verified partners of Mistral and his Kodestrali model, Sonet Claude 3.5 from Anthropik and Deepseek-R1 from Ollamy. However, each individual seller or developer can contribute to blocks and assistants to the hub.
The block here can mean models, which permit you to determine which model AI to use and where; rules to adapt the AI assistant; context define an external context supplier (e.g. jira or convergence); hint To pack pre -written monitors of the model for calling complex instructions; documents define documentation places (e.g. Angular or React); datawhich permit programmers to send development data to the predefined destination for analytical purposes; Or MCP serverswhich define the standard way to build and share tools for language models.
“Culture of contribution”
The idea of this recent center is that almost all users won’t require deep adaptations – they’ll only have to make minor corrections to coding assistants or blocks that exist already in Huba.
This raises the query: what is the encouragement to create adaptations and sharing them with the world? As it seems, this is what drives the Open Source community elsewhere. Many starting partners are the corporations themselves that create basic tools or models (e.g. Mistral and Anthropic), thanks to which the recent continuation center is an ideal place for curry for programmers.
What’s more, “Ethos Open Source” is at the basis of what he strives for. So if someone created any adaptations to use at work, why not share it with a wider community? Ultimately, the continuation is positioned as an antithesis of reserved assistants AI “Black Box”.
“This is a center for the entire ecosystem to gather and cooperate,” said Dunn. “Instead of all those who build their own AI assistant closed, what if we had an open architecture in which we can all cooperate to create components that people need to create adapted experiences for themselves?”
This is what Dunn calls the establishment of a “cultural culture”, with developers encouraged to experiment and create their very own adaptations, while generating value for everyone.
“Thanks to continuing 1.0, we enable this culture of contribution for programmers to create and share non -standard AI code assistants,” said Dunn. “This register will be a place of discovery in organizations and, among other things, which will grow in the classroom along with the evolution of blocks and open programming tools with AI-Przedznami.”
Then there is a data control aspect. On a more general “universal” platform, the supplier can distinguish a significant remark value how programmers operate on a large scale, and transfer the decision -making data back to the platform to improve things for everyone. This form of activity caused controversy for similar to Github Copilot, which were accused of kidnapping the exertions of tens of millions of Open Source programmers for their very own profits.
Continuing, the idea is that corporations have more control over what is happening to their data – they’ll share as much or as much as they need.
“When you use, continue, you can keep your data,” said Dunn. “As an organization, you can combine all your data for all programmers in one place. This is not possible in one size, the Black Box Code assistant, where their offer and strategy SaaS is to accept data and use them to improve them for everyone. “
Model business
This is still a relatively early days for continuing, but the startup claims that he cooperated with a handful of well-known corporations in the development phase Ionian(also an early continuation customer), in addition to Siemens and Morningstar.
While large corporations are very focused, Dunn claims that continuing is directed to programmers of all shapes and sizes, from freelancers and small teams to a range of enterprises. This indicates how to proceed money – its recent concentrations are sent by a free solo level, but organizations that need more control over their data pays to access additional administration, management and security tools.
“There is a lot of interest from larger organizations, but we have also seen everything for individual programmers, who simply wants a kind of adaptation for themselves. In such cases, I think that the solo level will be more than enough – said Dunn. “But when this freelancer or small team begins to grow and need some management, they’ll change into clients.”
Free solo levels send three levels of “visibility”. The developer’s contribution might be maintained on privacy, divided internally as a part of the team or fully made public. Indeed, the solo level might be technically used in the configuration of the team; She only lacks some functions that the team would normally require. A separate level of “teams” adds additional intelligent “many players” to the mix, with administrative controls to the governing all blocks and assistants-who have access to it.
Meanwhile, the company level increases data, security and management options, with more detailed control over what blocks, models, versions and suppliers are used.
“The administrator can also manage safety around the certificates in which the data go and receive the audit journal for WHO, what, when and where using programmers,” said Dunn.
Continue earlier raised $ 2.1 million after completing the Y Combinator at the end of 2023, and now raised one other $ 3 million of safes (financing with delayed capital allocation) run by VC focused on developers, Heavy.
Dunn says that almost all fresh money might be allocated to software engineering salaries and plan to “at least double” their current variety of guesthouses of 5.
“We use Open Source as an approach to distribution, as a result of which we maintain our very low costs – we do not have to capitalize almost as much as other competitors,” said Dunn.