Large language model (LLM) provider Cohere has unveiled its new agentic AI offering — North — a low-code platform that will allow enterprises to build and deploy agents across different business functions.
The offering will compete with Microsoft’s autonomous agents, Google’s Vertex AI agents, and Salesforce’s Agentforce.
Currently only being offered as part of an early access program, North will allow enterprises to build agents that help find relevant information across global knowledge repositories in multiple languages, conduct research & analysis, and perform complex tasks spanning various lines of business.
“This includes agents for core business functions like HR, finance, customer support, and IT that allow teams to execute faster and achieve more,” Cohere said in a statement.
The tech stack behind North
One of the core tenets of North is Cohere’s proprietary multimodal AI search and discovery framework — Compass.
Compass itself combines retrieval models such as Embed and Rerank, document parsing abilities, and a managed index.
While the retrieval models help with RAG applications, the document parsing ability pre-processes documents and supports PDF, PPT, DOCX, and XLSX formats.
The ability to support different formats is key, according to Aidan Gomez, CEO and co-founder of Cohere, as extracting information from fragmented data is essential for any AI system as it needs to surface insights for employees or end-users to take action on.
The managed index inside compass works like a managed service that manages the index or the vector database to improve performance and reduce latency, the company said.
Cohere said North will come with a security system that can imbibe an enterprise’s identity and access management rules and can be deployed in private cloud or on-premises.
This, according to Gomez, was one of the prominent demands among customer enterprises as they want to ensure that sensitive data is not leaked, making the offering “well-suited for regulated industries where companies simply cannot risk their proprietary data.”
Among the early customers of North is the Royal Bank of Canada, which is co-developing a North for Banking offering with Cohere.
Focus on ease of building and deploying agents
Cohere is positioning North for its ease of use for building and deploying agents, a growing concern among developers who are trying to build applications underpinned by generative AI.
A survey conducted by IBM involving developers across at least 1,000 enterprises revealed that though almost everyone is exploring how to use agents in their workflows, at least 31% are concerned about their trustworthiness.
Nearly 23% and 22% are concerned about cybersecurity threats and agents losing visibility into systems respectively, according to the survey.
Cohere isn’t the only one focusing on the ease of use of agentic platforms.
In December, Salesforce unveiled the next generation of its low-code agentic platform — Agentforce 2.0 — with an updated reasoning engine that offers the ability to build agents using natural language and new agentic skills that can perform more tasks without user intervention.
Another major upgrade was Agentforce’s integration with MuleSoft, designed to help enterprises reduce the time and complexity of building a new custom agent and integrating it into a workflow.
In the same vein, Cohere touts that North, which combines its proprietary AI search, LLMs, and agents, can be integrated “seamlessly” into any existing workflow out-of-the-box.
“AI agents created with North can quickly and easily connect to the workplace tools and applications that employees regularly use,” Gomez wrote in a blog post, adding that North can also be integrated with in-house applications.
However, he didn’t flesh out how exactly the entire integration process works. Microsoft, too, upgraded Copilot Studio — its agent-building platform — with the ability to allow enterprises to connect agents to third-party applications such as Salesforce, ServiceNow, and Zendesk.