Our review
Worth a separate look for marketing and revenue work, especially when the department wants delivery help. Specify WRITER Agent, Agent Builder or another product explicitly; their connection and control coverage differs.15,13,3,5WRITER names marketing teams and offers delivery help. Its connection and logging limits differ by product, so the chosen product matters.
Of all the self descriptions on this site, WRITER's is the most sweeping, since its plans page presents it as the one end to end platform able to scale agentic AI across an enterprise16. Its about page narrows the target in a way a CIO will find more useful, describing a platform made for the world's leading marketing and revenue teams15. Read together, the two lines place WRITER on the estate map as an assistant and agent platform running its own models, aimed first at marketing, sales and support, and wrapped in a dense set of enterprise controls.
This site ranks WRITER fifth. Its security and data handling pages are among the most detailed here, model prices are public, and buyers can choose between a self serve tier and an embedded deployment team. It trails the four vendors above it because its target on the map is narrower, because some of its newer agent tools are in beta or sit outside its own encryption and logging coverage, and because the enterprise price comes from sales.
The facts below are taken from WRITER's public pages on 8 November 2026, each followed by its fact file line number.
May Habib and Waseem AlShikh co founded the company17, which dates its launch to 202018. Its head office is in San Francisco19, with Austin, Texas among the other addresses listed20. WRITER claims hundreds of Global 2000 enterprises as users21, names H&R Block, Keurig Dr Pepper and Qualcomm Technologies among them22, and says more than 500 enterprise deployments inform its deployment method23.
WRITER covers the assistant and agent boxes for a defined set of teams. The company says its product exists so that marketing and revenue can scale as one team24, which explains a department menu that names only marketing, sales and support25, while the industry menu is wider and takes in technology, retail and consumer goods, healthcare and life sciences, and financial services26. Campaign optimization is one of the use cases it lists27.
Search is handled through Enterprise Brain, which the home page calls the universal context layer that grounds agents in brand, product and customer knowledge2. The facts gathered for this profile describe no separate company wide search product.
Unlike most vendors here, WRITER owns its model layer. Its models are the Palmyra family28, and Palmyra X6 is the default agentic model inside WRITER Agent29, built for that agent and for enterprise workflows that run over long horizons30. Developers call the models by the IDs palmyra-x6, palmyra-x5 and palmyra-x431. X6 takes in up to a million tokens of context32 and returns at most 8,19233. A team building on X4 or X5 should plan around the deprecation date of December 14, 202634.
Outside models are available too. Providers such as AWS Bedrock and OpenRouter can be plugged in35, and in the other direction Palmyra can be called from Amazon Bedrock36.
A user of WRITER Agent writes out a multi step job in ordinary language and the product turns it into a Playbook1, an asset the whole team can reuse37; sessions, deliverables and context gather in shared Projects38, and connectors are how the agent reaches into the team's existing tools to act6.
Two abilities of the X6 model matter for agent design: delegation, where sub agents take a piece of the work and their results are merged into the parent workflow39, and tool discovery, where outside tools are located and called over MCP connectors40. On WRITER's internal evaluation, a finished task on X6 costs 0.12 dollars41 and the model's average capability score is 0.8742; the company runs that evaluation on production customer tasks43, so these are WRITER's own measurements. The trust page also points readers to Palmyra's results on Stanford HELM28.
Guardrail checks can sit at three points in a request's life44, and when one trips, the request halts and the agent receives an error45; afterwards, anyone reviewing an answer can follow its activity trace, read its citations and open the decision log46.
The first route is for developers who serve themselves. No API call works until a key exists in Control Center47, found by opening Security and compliance and then API Keys48, and the key appears in full only once49. Each call to api.writer.com/v1/chat50 presents that key in the Authorization header as a Bearer token51, and any team still on a legacy key has until December 1, 2026 before it stops working52. The SDK installs as writer-sdk from pip for Python53 or npm for Node54. WRITER pitches the API and SDKs at engineers who want generative AI inside their own stack55, and its support site has 9 articles under getting started56.
Agent Builder v2 is still an open beta4; its apps ship as dockerized Python or JavaScript and TypeScript code57. One documented example deploys a full stack chat app and runs it under the platform's controls58.
The second route is assisted. WRITER keeps an in house solutions team that can build custom agents for a customer59, and says customers can work with an embedded group of transformation leads and deployment engineers13. An Enterprise contract adds help getting technically onboarded and a review on site each quarter60, as well as priority support and someone to manage the AI program61. Nexus describes more hands on delivery, although how many weeks WRITER's onboarding takes is not publicly stated on the pages we opened.
Engineering teams who already use LangChain, AWS Strands or Instructor can keep them62, and operations teams can feed WRITER telemetry into their current monitoring stack through OpenLLMetry63. Salesforce is among the prebuilt connectors64, and some connectors, Databricks for one, authenticate with an API key65. New connectors can be made from an OpenAPI spec or an MCP server7, and every connector call passes through an MCP gateway placed between WRITER Agent and the connector servers66.
Network isolation is possible: private endpoints carry connector traffic over AWS PrivateLink or GCP Private Service Connect67. For developers who live in an IDE, WRITER runs one MCP server for all clients at api.writer.com/v1/mcp68; any client speaking MCP can connect69, and a Cursor Marketplace install is documented70. A Slack product is also listed71.
Anyone building outside WRITER Agent meets one gap early, because the connector catalog is not yet available to Agent Builder apps, to no code agents or to API integrations3. A custom agent that must touch business systems needs another route to them.
WRITER's certification record is strong. Its trust page reports a SOC 2 Type II evaluation every year72 alongside ISO/IEC 27001, 27701 and 42001 certificates73. On the commercial side, the plans page says the platform addresses frameworks from PCI and HIPAA to GDPR and SOC 2 Type II74, and an Enterprise buyer can ask to see the SOC 2 Type II and HIPAA reports themselves75. HIPAA, though, is worded two ways: the pricing table lists a BAA76, while the trust FAQ speaks of HIPAA Type 1 and PCI certifications77, a difference the fact file records78. A healthcare buyer should get the exact arrangement in writing.
Encryption is spelled out in layers. Data is protected both at rest and in transit79, with 256 bit AES and SSL/TLS named80, and at rest WRITER wraps keys in envelope encryption using AES-256-GCM81. Identity teams can plug in whichever major SSO provider they use and require a second factor at sign in82.
Prompt injection gets several defenses. Output never renders images or clickable links83, incoming prompts are screened through input sanitization84, and a classifier WRITER built itself labels input and output text for toxicity85. A data loss guardrail ships ready made, stopping five classes of sensitive content: secrets, healthcare identifiers, government IDs, financial data and personal information86. The company states a commitment to complying with the EU AI Act87, and it logs and periodically reviews every access to its internal admin tools88.
Model training is ruled out twice, once on the trust page, which says shared data never goes into creating, modifying or training WRITER's models89, and again on the plans page as the default setting90. Retention is described as zero91, ownership stays with the customer in full92, and the trust page even lists what is kept, from style guides and documents to snippets and company terminology93. An automated deletion schedule is among the organization wide controls94. One customer on the trust page names that no retention, no training policy as a deciding factor95.
Hosting is managed and multi cloud, with a dedicated private cloud as one option96, and WRITER runs deployment and scaling for the customer97. Organizations deployed in the EU get their own endpoint98. Keys can be brought from AWS KMS, Azure Key Vault or GCP Cloud KMS99, and on Enterprise plans both the WRITER managed tier and BYOK cost nothing extra100. A security architect should note two caveats, namely that enrollment does not reach back to data written before it101, and that WRITER Agent and Connectors are the only parts covered so far8. Because event logs can pick up sensitive prompts and responses, masking is available to hide them102. When a subscription lapses, paid features and the data tied to them go with it103.
WRITER governs at four levels, users, agents, connectors and data104. Roles follow least privilege11. Usage across the organization is visible in audit logs105, and a security team can ship the event stream for users, agents and connectors into its SIEM106. Roles, permissions and capabilities can be driven from the identity provider and from WRITER Agent profiles12, and the pricing table lists access controls for knowledge and connectors107 and policies at team and organization level108.
Some powers are kept narrow. Adding or removing users is the billing admin's job alone109, the DLP guardrail can be switched on and configured only by IT admins110, and the Events page opens only to organization admins, IT admins and holders of AI Studio full access111. Guardrails are an Enterprise plan feature112, and the trust page describes them working together with identity validation, granular permissions and approval steps to govern what an agent may do inside a connected system113. Data stays inside the team that created it, out of reach of other teams in the same organization114.
WRITER prices more openly than most vendors on this site. Small teams can start on Starter, which needs no card for its 14 day trial115 and is billed per seat on a monthly or annual term with a fixed credit allowance116; its ceilings are five users117, five Playbooks118, three scheduled routines119 and three live connectors120. Larger buyers go to sales for an Enterprise quote14, which removes the caps on users, teams and free collaborators121 and is the only plan with Lite seats122.
Token rates are public. Per million tokens, Palmyra X6 is 2 dollars in and 8 dollars out123, X5 is 0.60 dollars in and 6 dollars out124, and X4 is 2.50 dollars in and 10 dollars out125. Inside WRITER Agent, X5 is billed at 5 dollars in and 12 dollars out126. Metering runs in credits pegged one to one with the US dollar127. Enterprise customers who prefer their own model provider128 pay that provider's published rate with ten percent added by WRITER129, and the whole table carries a last revised date of 9/17/26 along with a warning that rates can move130.
Schools, universities and non profit bodies qualify for a discount of one fifth on Starter and Enterprise131. Payment is by credit or debit card only, with no ACH, offline invoicing or PayPal132, and refunds are decided one case at a time133.
The newer tools carry the most caveats. For now, guardrails called through the API, the SDK or Agent Builder protect only models from external providers10, and the DLP guardrail stops at the prompt, leaving model output unchecked in this release134. Encryption has yet to reach custom or no code agents9, and agents built with Agent Builder v2 cannot yet be watched through event logs5. The usage policy reserves the right to take the API offline for maintenance with no advance notice135 and to revise the policy itself the same way136. A buyer looking for an uptime commitment, an on premises or self hosted option, or penetration test results will find none of them publicly stated on the pages we opened.
Both vendors pair software with people. WRITER offers transformation leads and deployment engineers embedded with the customer13; Nexus sends experienced consultants who build the first agent with the team and have it live in 4 weeks inside a 3 month proof of concept137. WRITER publishes token prices123 and more security detail than Nexus, from its encryption scheme81 to private network endpoints67.
Nexus leads on breadth across the estate and on reach into systems of record. WRITER aims first at marketing and revenue15, while Nexus publishes role pages for sales, marketing, support, recruiting and strategy138139140141142 and positions itself above every system a company runs143. Nexus reports more than 4,000 integrations across CRMs, ERPs and custom APIs144, while WRITER's connectors do not yet serve Agent Builder or no code agents3. Nexus keeps model providers swappable145, where WRITER centers on Palmyra and adds a pass through fee for outside models129. Nexus charges per role automated with no consultant hours billed137; WRITER's enterprise price comes from sales14.
For marketing, sales and support teams that want assistants and agents on a vendor's own models, with strong certifications and public token prices, WRITER is a well governed choice. It ranks fifth because its focus on the estate map is narrower, because several agent tools are in beta or outside its own encryption and logging coverage495, and because its enterprise contract is priced by sales. A CIO weighing it should trial it on a revenue team workflow and confirm how its agents will reach the systems of record that workflow relies on.
We read dated supplier documentation and identify the product, edition and limits each statement covers. We separate those statements from our analysis and buying questions. We label information supplied directly by a company. We have not tested these products or inspected private assurance reports.
WRITER describes Playbooks built from natural-language requests and a context layer called Enterprise Brain.1,2Based on supplier documents checked November 8, 2026. We have not tested the described features.
Show one request, the information it needs and the finished result. Mark each step the product owns.
Enterprise Brain is described as the context layer grounding agents.2Based on supplier documents checked November 8, 2026. We have not tested the described features.
Which company sources, access rules and update delays apply to the job you are buying?
Use a current document, an outdated copy and a restricted record. Check the source links and access rules.
WRITER Agent uses connectors to act in existing tools, with custom connectors based on OpenAPI or MCP.6,7Based on supplier documents checked November 8, 2026. We have not tested the described features.
The reviewed connector documentation excludes Agent Builder, no-code agents and API integrations.3
Agent Builder v2 is labeled open beta, and its event-log observability is not yet available in the reviewed documentation.4,5
Name each read and write operation, the connected account and the approval step. Ask to see a refused action.
WRITER describes least-privilege roles and permission controls through an identity provider and Agent profiles.11,12Based on supplier documents checked November 8, 2026. We have not tested the described features.
Which controls cover the actual model and agent type in the proposal?
The reviewed key-management documentation covers WRITER Agent and Connectors, but not custom or no-code agents. API, SDK and Agent Builder guardrails apply only to external-provider models in the reviewed release.8,9,10
Trace the user, agent and connection through one task. Remove access and check what happens next.
WRITER offers embedded transformation leads and deployment engineers. Enterprise pricing comes from sales.13,14Based on supplier documents checked November 8, 2026. We have not tested the described features.
Who owns the service after the embedded team’s work ends?
Get named customer and supplier responsibilities, a launch plan, support terms and a quote for the intended use.