Writer

San Francisco based, $200M Series C at a $1.9B valuation in November 2024, and now positioned as an agent platform rather than a writing tool. The stack has four parts: the Palmyra models, a Knowledge Graph that indexes connected enterprise sources for grounded retrieval, AI HQ for building and orchestrating agents, and AI Studio for assembling apps with little code. Skills, added in March 2026, let a team encode its own methods and quality standards as reusable components that agents pull in.

Palmyra X6 arrived on 13 August 2026 with a 1M token context window and claims of 52 percent lower cost per agent run and 48 percent faster output than the previous generation, alongside upgrades to the agent harness. Worth knowing that it is post-trained on GLM-5.2, an open-weight base from Z.ai in Beijing. That is not a criticism of the result, post-training on a strong open base is a rational engineering decision and it is how the cost claims become plausible, but it does change how you read the model as a moat. Writer has also published open-weight domain models such as Palmyra-Med and Palmyra-Fin under its own licence, which are not OSI-approved and carry usage restrictions.

Who this is for is narrow and it is worth being direct about it. A regulated company with a content operation, brand and terminology rules that legal cares about, and data spread across systems it does not want to expose to a general API vendor. For that buyer the governance and the audit trail justify the seat price. For an engineer or a small technical team writing docs and posts, the Starter tier is expensive for what it gives and a general model with your own retrieval will go further for less.