WRITER

WRITER on cost: a public rate card behind a sales led seat price

WRITER publishes more of its usage economics than almost any vendor in this catalog. Model prices per million tokens, a credit that equals one dollar, a pass through fee for outside models and per page prices for knowledge extraction all appear on its developer pages^1,2,3,4. Its Enterprise seat price, by contrast, is available only through sales⁵. A finance team at a 1,000 employee company can therefore model the variable part of a WRITER bill in some detail and must request the fixed part.

WRITER's plans page claims that no other platform covers the full journey of scaling agentic AI in a large company⁶, and the about page aims the product at marketing and revenue teams⁷. That focus matters to the cost model, because it suggests the natural seat population inside a 1,000 person company is the marketing, sales and support staff⁸, which may be a fraction of total headcount. This profile places WRITER fifth.

What the platform does

WRITER Agent builds Playbooks, multi step workflows created from natural language⁹, and Playbooks are reusable workflows shared across a team¹⁰. Projects are shared workspaces for sessions, deliverables and context¹¹. Connectors let agents take action securely in the tools a team already uses¹². A context layer called Enterprise Brain feeds agents what the company knows about its brand, products and customers¹³. WRITER builds its own Palmyra models¹⁴. The newest, X6, targets WRITER Agent and long running enterprise workflows¹⁵. It spawns sub agents and merges what they return¹⁶, and it calls external tools through MCP connectors¹⁷.

Reuse has a cost meaning here. A Playbook built once and shared across a team spreads its build cost over every run. Starter's limit of 5 Playbooks¹⁸ is one reason a larger company outgrows it quickly.

The two plans

Starter is the self serve plan. Seats are paid by the month or the year, each plan carrying a fixed credit allowance¹⁹. Limits are tight. The user cap is 5²⁰, the Playbook cap is also 5¹⁸, and routines and connectors stop at 3 each^21,22. It comes with a 14 day free trial that needs no credit card²³. We did not record the Starter seat price when the pages were checked, so it is not quoted here.

Enterprise prices come from the sales team⁵, and the plan places no limit on users, teams or unpaid collaborators²⁴. Lite seats are only available on Enterprise plans²⁵. The developer pricing page repeats that custom pricing comes through sales²⁶.

For a 1,000 employee company, Starter's five user limit means Enterprise is the only fit. The availability of free collaborators and Lite seats on Enterprise^24,25 suggests that not every user has to be priced as a full seat, which can lower the total; how Lite seats are priced is not publicly stated on the pages we opened.

The usage rate card

Usage is metered in credits, with 1 credit equal to 1 USD². Published model rates per million tokens are:

1. Palmyra X6: $2.00 input and $8.00 output¹. 2. Palmyra X5: $0.60 input and $6.00 output²⁷. 3. Palmyra X4: $2.50 input and $10.00 output²⁸.

X5 costs more through WRITER Agent, where a million input tokens are billed at $5.00 and a million output tokens at $12.00²⁹. Rates are subject to change and were last updated on 9/17/26³⁰.

For the Knowledge Graph, each page extracted costs $0.00015⁴ and each page fetched from the web costs $0.12³¹. On our arithmetic, extracting one million pages costs $150.

WRITER's internal tests put X6 at $0.12 for each task completed³². Those tests were run by WRITER on tasks drawn from its customers' production work³³. A buyer should treat the per task figure as a starting estimate and measure its own.

Bringing your own model

A customer using its own model pays that provider's list rates with a 10 percent WRITER fee on top³; the option is limited to Enterprise and covers both the API and WRITER Agent³⁴. OpenRouter and AWS Bedrock are among the outside model sources that can be plugged in³⁵. For a company with negotiated model rates elsewhere, the 10 percent fee is a known, modest overhead.

Model lifecycle and its cost effect

WRITER will retire X4 and X5 on December 14, 2026³⁶. X6 is now the default model in WRITER Agent³⁷, with a context window of 1M tokens³⁸ and output capped at 8,192 tokens³⁹. Since X6 is priced differently from X5^1,27, a workload moved from X5 to X6 will change cost. Finance should reprice any forecast built on X5 rates.

Legacy API keys stop working on December 1, 2026⁴⁰, a date that matters to teams with existing integrations.

Implementation support

WRITER's pages we opened list implementation support as part of a plan. Enterprise adds technical onboarding and quarterly onsite reviews⁴¹, along with priority support and AI program management⁴². WRITER's home page promises deployment engineers and transformation leads working inside the customer's team⁴³, following a method drawn from upward of 500 enterprise rollouts⁴⁴. Its own solutions staff can also build custom agents for a customer⁴⁵. Whether that solutions work is included in the Enterprise price or billed separately is not publicly stated on the pages we opened.

For self built work, the developer path is documented. Developers start by generating a key in Control Center⁴⁶. SDKs install with pip or npm^47,48. Agent Builder v2 remains an open beta⁴⁹; apps built with it ship as Docker containers written in Python, JavaScript or TypeScript⁵⁰. The support knowledge base lists 9 articles in its getting started section⁵¹. An implementation timeline is not publicly stated.

Integrations and their cost

Developers can work through LangChain, Instructor or AWS Strands⁵² and send telemetry to their current monitoring stack by way of OpenLLMetry⁵³. Palmyra models can be invoked on Amazon Bedrock⁵⁴, which may let a company with AWS commitments draw down existing spend. Prebuilt connectors include Salesforce⁵⁵; some use API key authentication, for example Databricks⁵⁶. A custom connector starts from an MCP server or an OpenAPI spec⁵⁷; traffic between WRITER Agent and connector servers passes through an MCP gateway⁵⁸. For private connectivity, WRITER supports Private Service Connect on GCP and PrivateLink on AWS⁵⁹. The WRITER MCP server is hosted at a single address⁶⁰, works with any MCP compatible client⁶¹ and has a documented setup for Cursor⁶². A WRITER for Slack product is listed⁶³.

Procurement details

Two items on the plans page deserve a check against accounts payable practice. WRITER takes card payments; it currently accepts neither PayPal nor ACH, and it does not issue offline invoices⁶⁴. Whether Enterprise contracts follow the same rule is not publicly stated on the pages we opened. Refunds are approved case by case⁶⁵. Users can be added or removed by the billing admin alone⁶⁶, which leaves finance one point of control over the seat count.

The exit terms are explicit. When a subscription expires, the customer loses access to paid features and all data associated with them⁶⁷. Cancellation runs through Org settings, then Billing, then Cancel plan⁶⁸. A buyer should plan data export before any non renewal.

Schools, universities and non profits receive 20 percent off both plans⁶⁹.

Security and compliance

WRITER's compliance evidence is strong. Each year WRITER is evaluated for SOC 2 Type II⁷⁰, and it holds three ISO/IEC certificates: 42001, 27701 and 27001⁷¹. Its plans page names PCI, HIPAA, GDPR and SOC 2 Type II among the frameworks covered⁷², and Enterprise customers can ask for the HIPAA and SOC 2 reports⁷³. On HIPAA the pages differ⁷⁴. The pricing table lists a BAA⁷⁵; the trust page FAQ instead mentions certification for HIPAA Type 1⁷⁶.

Stored and moving data are both encrypted⁷⁷, with AES at 256 bits and SSL/TLS⁷⁸. Enterprise customers can let WRITER manage keys or supply their own at no extra charge⁷⁹, using the key services of Azure, GCP or AWS⁸⁰. Only data created after enrollment is encrypted this way⁸¹, and the scheme does not yet extend to custom agents or no code agents⁸². All major SSO providers and multi factor authentication are supported⁸³.

WRITER says customer data never goes into building or training its models⁸⁴, and it describes its retention stance as zero retention⁸⁵. Deployment is managed and multi cloud, including dedicated private cloud⁸⁶. An EU endpoint is available for organizations deployed in the EU⁸⁷. Uptime service levels and on premises deployment are not publicly stated.

Governance and spend control

Controls reach four levels: users, agents, connectors and data⁸⁸. Roles and permissions can be controlled from the customer's identity provider and WRITER Agent profiles⁸⁹. Audit events about users, agents and connectors can flow into a SIEM⁹⁰. Team and org level policies⁹¹ and knowledge and connector access controls⁹² are listed on the pricing table. Guardrails come with Enterprise⁹³, and the DLP guardrail can be switched on by IT admins alone⁹⁴.

Some limits apply. In this release DLP checks inputs and skips what the model returns⁹⁵. Agents made in Agent Builder v2 do not yet appear in event log observability⁹⁶. For now, connectors cannot be used from Agent Builder, from no code agents or through API integrations⁹⁷. Specific spend caps of the kind Copilot Studio offers are not publicly stated on the pages we opened.

Further limits from WRITER's own pages

Guardrails used with the API, SDK and Agent Builder apply only to external provider models for now⁹⁸. Its usage policy allows unannounced API downtime for maintenance⁹⁹ and unannounced changes to the policy itself¹⁰⁰. A transparency page mentions an Apache 2.0 license for the models¹⁰¹; we have not checked that against the product pages. Prompts and answers recorded in event logs may hold sensitive data, which masking can conceal¹⁰². Admin tool access at WRITER is logged and periodically reviewed¹⁰³, and one team's inputs are not accessible to another team in the same organization¹⁰⁴.

Where it fits on cost

WRITER suits a company whose AI spend will concentrate in marketing, sales and support^105,8, which wants published model rates to forecast usage¹, and which values onboarding support inside an Enterprise contract⁴¹. Named customers include H&R Block, Keurig Dr. Pepper and Qualcomm Technologies¹⁰⁶, and the about page speaks of hundreds of Global 2000 customers¹⁰⁷.

It is harder to price for a company that wants one product for all 1,000 employees across every function, since the seat price is by quote⁵ and the product's stated focus is narrower⁷.

Why it ranks fifth

WRITER ranks fifth because its usage side is the most transparent here after Copilot Studio's packs^1,2,3, and its Enterprise plan includes onboarding support⁴¹. It sits below Glean because Glean's public deployment record at large scale¹⁰⁸ gives a 1,000 employee buyer more evidence on the value side, and because WRITER's card only payment statement and data loss at expiry^64,67 add procurement items to resolve. It sits below Nexus because Nexus prices per role automated, is not tied to headcount and bills no per diem or consultant time^109,110, while WRITER's Enterprise seats and its solutions team pricing are not publicly stated on the pages we opened.