Writer

What Writer is

Writer is an enterprise AI company. Rather than selling a single general-purpose chatbot, it sells a set of tools aimed at marketing and revenue teams: on-brand agents, an Enterprise Brain built on a company's brand DNA, and governance controls that sit around them ^[1]. The pitch is that large organisations want generative AI that writes in their own voice and stays inside their own rules, rather than a general model answering from general knowledge.

Origins

Writer launched the company in 2020, according to its own history of the firm ^[1]. It was co-founded by May Habib, who is chief executive, and Waseem AlShikh, who is chief technology officer ^[1]. The two met over a tweet about machine translation and later ran a company that localised enterprise content at scale ^[1]. That background — moving text between languages and markets for big organisations — is close to the problem Writer now sells against: getting a model to produce text in one particular company's voice.

By its own account, the team began experimenting with transformers in 2018, and a model-written haiku read out at an all-hands in 2019 led to the idea for the company ^[1]. Transformers are the neural-network architecture behind most modern language models. They process a sequence of text by letting each token weigh its relationship to the other tokens around it — the mechanism known as self-attention — which makes it practical to train one model on very large amounts of text. A short poem produced by a small model was enough to suggest that the same machinery could generate usable marketing copy.

What it sells

The company names H&R Block, Keurig Dr. Pepper and Qualcomm Technologies among its customers, and says it serves hundreds of Global 2000 enterprises ^[1]. Its headquarters are in San Francisco, with offices in New York, London, Chicago and Austin ^[1].

The product set is organised around three pieces. Agents do the writing and the task work; the Enterprise Brain is described as being built on brand DNA, meaning the model is grounded in a company's own material and style; and the governance controls are there to keep outputs compliant and auditable. Selling all three together is a common shape for enterprise AI: the model alone is rarely the hard part, and the surrounding work of connecting it to a company's content and rules is where deployments succeed or stall.

Open models

In September 2025 Writer published the Palmyra-mini family as open models, with the base model on Hugging Face on 5 September and two thinking variants on 10 September ^[2]. All three are released under Apache 2.0, are English-only, and list a 131,072-token context window ^[2][3].

Apache 2.0 is a permissive licence: anyone can use, modify and redistribute the weights, including commercially, provided they keep the notices. A context window is the amount of text a model can hold in view at once; a token is a chunk of text roughly the size of a short word or part of a word, so 131,072 tokens corresponds to a long document rather than a paragraph.

The family sits between 1.5 and 1.7 billion parameters, and GGUF and MLX builds are published in the same Hugging Face organisation ^[2][3]. Parameters are the learned numbers that make up a model, and the count is a rough proxy for capability and cost; at this size a model is small enough to run locally rather than only through a hosted API. GGUF and MLX are packaging formats used to run models on ordinary hardware.

Where it sits

Writer belongs to the group of companies selling generative AI into large organisations rather than to consumers, competing on fit with a company's own content and on the controls around it. Publishing small open models alongside a commercial platform is a second, narrower role: it puts runnable versions of its models in public, where outside developers can download, test and build on them.