WRITER

WRITER for FCA-regulated firms: an assessment

On its plans page, WRITER claims to be the one end to end platform on which an enterprise can scale agentic AI⁹. The about page narrows that down: WRITER is built for the marketing and revenue teams of leading companies¹. For an FCA-regulated firm, that combination is interesting. Financial promotions, customer communications and sales material are relevant workflows to evaluate here. A firm considering them should test how the marketing and revenue features support its own approval, evidence and review requirements.

WRITER is placed sixth in this ranking. It publishes one of the most complete security and data handling accounts of any contender, offers an embedded delivery team⁸ and lists financial services among its industries¹⁰. It sits behind the five contenders above because several of its controls carry stated gaps in the current release, because its focus is marketing and revenue work, and because a number of developer features are in beta.

The profile uses WRITER's own public pages, checked on 8 November 2026. A source number accompanies each statement.

The company

WRITER says it launched in 2020¹¹, founded by May Habib and Waseem AlShikh¹². Its headquarters is in San Francisco¹³, with other offices including Austin, Texas¹⁴. Customers it names include Qualcomm Technologies, H&R Block and Keurig Dr. Pepper¹⁵, and it counts hundreds of Global 2000 companies among its users¹⁶. It cites more than 500 enterprise deployments behind its deployment methodology¹⁷.

How the platform works

With WRITER Agent, a user describes a multi step workflow in natural language and receives a Playbook¹⁸. A Playbook can then be reused and shared by the whole team¹⁹. Projects give the team a common workspace that holds sessions, deliverables and context²⁰. Connectors let agents act in the tools a team already uses²¹. A context layer called Enterprise Brain gives agents their grounding in the firm's brand, products and customers².

For a regulated marketing team, shared and reusable Playbooks are a governance asset. A workflow approved once by compliance can be reused, which reduces the variation that comes from each person prompting a model in their own way.

WRITER also builds its own models. Its language models go by the name Palmyra²²; WRITER Agent runs by default on Palmyra X6²³, which the company pitches at long running enterprise workflows²⁴. Through the API, the three models are called palmyra-x6, palmyra-x5 and palmyra-x4²⁵. Palmyra X6 has a 1 million token context window²⁶ and a maximum output of 8,192 tokens²⁷. X6 can hand tasks to sub agents and fold their results back into the main flow²⁸, and it reaches outside tools through MCP connectors²⁹. The older X4 and X5 models are due to be retired on 14 December 2026³⁰. WRITER reports an internal average capability score of 0.87 for X6³¹ and a cost of 12 cents per finished task on its own evaluation³²; both are the company's own measures on production customer tasks³³.

Owning the model has a consequence for a regulated firm. Its model risk review will need to look at Palmyra itself as well as the application built on it, so it should ask what documentation WRITER provides on the model's testing and change history.

Guardrails and explainability

WRITER's guardrails can check content at three points in a request: before the call, after it and during it³⁴. When a guardrail detects a violation, the request halts and an error returns to the agent³⁵. Its data loss prevention guardrail stops five kinds of content: personal details, financial data, government identifiers, healthcare identifiers and secrets³⁶. Only IT administrators can enable and configure it³⁷. Guardrails come with Enterprise plans³⁸.

Explainability is addressed directly. To explain an answer, WRITER offers activity traces, cited sources and decision logs⁶; to watch behaviour over time, it adds performance reports to those traces³⁹. For a firm that must be able to reconstruct why a customer received a particular piece of content, those features are relevant.

There are stated gaps. In the current release, model output passes the DLP guardrail uninspected⁴⁰. For now, when guardrails are used through the API, SDK or Agent Builder, they cover external provider models only³. A firm should map which of its intended uses fall inside guardrail coverage before relying on it.

WRITER also describes prompt injection defences, including input sanitisation in which incoming prompts are screened⁴¹. The platform prevents rendering of images or clickable links in output⁴², and the company runs its own classifier on input and output text for toxicity⁴³.

Governance and permissions

Governance is offered at user, agent, connector and data level⁴⁴. Role based permissions and least privilege access help ensure each action is authorised⁷. Roles, permissions and capabilities can be controlled from the firm's identity provider and WRITER Agent profiles⁴⁵. The plans table adds access controls over knowledge and connectors⁴⁶ and policies at team and organisation level⁴⁷.

Approval is part of the published design. WRITER says the way agents act on connected systems is shaped by guardrails, identity validation, granular permissions and approval flows⁴. For customer facing content, approval flows are the control a compliance team will want to see configured.

Audit is well covered. Audit logs show how WRITER is used across the organisation⁴⁸. Events covering users, agents and connectors can be sent on to a SIEM⁴⁹. The Events page is restricted to organisation administrators, IT administrators and AI Studio full access roles⁵⁰. Because prompts and responses may contain sensitive material, the logs can mask it⁵¹. Observability for Agent Builder v2 agents is not yet available⁵.

Data separation inside a firm is stated. Inputs and data from one team are not used by or accessible to another team in the same organisation⁵². Internally, WRITER restricts access to user data to job functions that need it⁵³, and all access to internal administration tools is logged and periodically reviewed⁵⁴. Only the billing administrator can add or remove users⁵⁵.

Security and certifications

WRITER's trust page records an annual SOC 2 Type II evaluation⁵⁶ and three ISO/IEC certifications, 27001, 27701 and 42001⁵⁷. The plans page names four frameworks the platform addresses: SOC 2 Type II, GDPR, PCI and HIPAA⁵⁸. Enterprise customers can ask for the SOC 2 Type II and HIPAA reports⁵⁹. WRITER's home page says it meets the compliance standards of regulated industries⁶⁰, and its trust page commits the company to compliance with the EU AI Act⁶¹.

HIPAA is described differently across pages, with a BAA on the pricing page⁶² and HIPAA Type 1 certifications in the trust FAQ^63,64. That inconsistency is minor for a UK financial firm, but it is a reminder to rely on the reports themselves.

Encryption is described at several levels. Encryption covers data both stored and moving⁶⁵, with 256 bit AES and SSL/TLS named on the plans page⁶⁶ and envelope encryption under AES 256 GCM for stored data⁶⁷. Sign in works with every major SSO provider and with multi factor authentication⁶⁸. Firms that want to hold their own keys can use AWS KMS, Azure Key Vault or GCP Cloud KMS⁶⁹, and Enterprise plans include both this option and WRITER managed keys without extra charge⁷⁰. Data is grouped into scopes that each have a separate encryption key⁷¹, with the settings controlled by organisation administrators in Control Center⁷².

Encryption coverage has limits. It currently covers WRITER Agent and Connectors⁷³; custom agents and no code agents are not yet covered⁷⁴. Enabling encryption applies only to data created after enrolment⁷⁵.

Data handling and location

WRITER says the data customers share is never used to create, change or train its models⁷⁶, and its plans page repeats that customer data is not used for training by default⁷⁷. It takes a zero data retention approach⁷⁸. Ownership of the data stays entirely with the customer⁷⁹. Organisation wide controls include an automated data deletion schedule⁸⁰. One customer testimonial says the transparent policy on not retaining or training on data was a key factor in its choice⁸¹.

Hosting is managed by WRITER across more than one cloud, with a dedicated private cloud available⁸², and WRITER also handles deployment and scaling⁸³. An EU endpoint is available for organisations deployed in the EU⁸⁴. Connectors can reach private systems through AWS PrivateLink or GCP Private Service Connect⁸⁵. On the pages we opened, an uptime commitment and any on premises or self hosted option are not publicly stated.

For a UK firm, the EU endpoint⁸⁴ and the dedicated private cloud option⁸² are useful starting points. A firm that needs UK processing specifically should ask.

Delivery support

WRITER is one of the few contenders that describes hands on help. Customers, it says, are joined by an embedded team made up of transformation leads and deployment engineers⁸. An in house solutions team can build custom agents⁸⁶. Enterprise plans include technical onboarding support and quarterly onsite reviews⁸⁷, plus priority enterprise support and AI programme management⁸⁸. The support knowledge base has 9 articles in its getting started section⁸⁹. How long a WRITER implementation takes is not publicly stated on the pages we opened.

This brings WRITER closer to Nexus on delivery than most contenders. Two Nexus commitments open the gap between them here: an agent working by week 4 of the proof of concept, and pricing on the basis of each role automated⁹⁰.

Developer route

API keys are created in Control Center under Security and compliance^91,92. Each key is displayed in full only once and cannot be shown again afterwards⁹³. Calls use a Bearer token⁹⁴ to endpoints such as the chat endpoint⁹⁵. Older API keys will be switched off on 1 December 2026⁹⁶. Python and Node SDKs are installed with pip and npm^97,98. Agent Builder v2 is in open beta, with features that may change⁹⁹, and its apps are packaged as dockerised Python or JavaScript and TypeScript applications¹⁰⁰.

Integrations include LangChain, AWS Strands and Instructor¹⁰¹, telemetry export through OpenLLMetry¹⁰², Palmyra on Amazon Bedrock¹⁰³ and external models from AWS Bedrock and OpenRouter¹⁰⁴. Salesforce¹⁰⁵ and Databricks¹⁰⁶ are among the ready made connectors, and a firm can build its own from an OpenAPI specification or an MCP server¹⁰⁷. Between WRITER Agent and the connector servers sits an MCP gateway¹⁰⁸. Agent Builder, no code agents and API integrations cannot use connectors at present¹⁰⁹. A WRITER MCP server works with any compatible client^110,111, and WRITER for Slack is listed¹¹².

Pricing and commercial terms

Starter has a 14 day free trial with no credit card¹¹³, up to 5 users¹¹⁴, up to 5 Playbooks¹¹⁵, up to 3 scheduled routines¹¹⁶ and up to 3 active connectors¹¹⁷. It is bought per seat, monthly or annually, with a fixed credit limit¹¹⁸. For Enterprise, a price comes from the sales team^119,120, and the plan allows whatever number of users, teams and free collaborators the firm needs¹²¹. Lite seats are only on Enterprise¹²².

Model prices are published. Palmyra X6 costs 2 dollars input and 8 dollars output per million tokens¹²³. Usage is metered in credits, with one credit equal to one dollar¹²⁴. If a firm brings its own model, it pays the provider's rate with a 10 percent WRITER fee added¹²⁵. Rates were last updated in September 2026 and are subject to change¹²⁶.

Two commercial terms deserve a regulated procurement team's attention. Payment is by credit or debit card; ACH, offline invoicing and PayPal are not supported¹²⁷. And when a subscription expires, access to paid features and all associated data ends¹²⁸. A firm should agree export and exit arrangements in the contract. Refunds are at the company's discretion¹²⁹.

Limits to weigh

WRITER's stated limits are specific. The DLP guardrail does not inspect output⁴⁰. API, SDK and Agent Builder guardrails apply only to external models for now³. Encryption does not yet cover custom or no code agents⁷⁴. Observability is not yet available for Agent Builder v2⁵. WRITER reserves the right to take the API down for maintenance without warning¹³⁰ and to revise its usage policy without warning¹³¹. On the pages we opened, four things a regulated buyer might expect are not publicly stated: an uptime SLA, on premises deployment, penetration test results and an implementation timeline.

Verdict

WRITER gives an FCA-regulated firm a strong account of security and data handling: SOC 2 Type II, ISO 27001, 27701 and 42001^56,57, zero data retention⁷⁸, bring your own key⁶⁹, an EU endpoint⁸⁴, approval flows⁴ and SIEM export of audit events⁴⁹. Its embedded delivery team⁸ is a genuine strength.

It ranks sixth because its published focus is marketing and revenue^1,132, because several controls have stated gaps in the current release^3,40,74,5, and because commercial terms such as card only payment¹²⁷ will need negotiation in a large firm. Nexus is placed first on a stated time to first agent, a release gate tied to sign off and a commercial model priced per role. Confirm the current release limitations and enterprise terms before selecting it for the proposed communications workflow.