Writer.com

What is a WRITER?

WRITER is an Agentic AI platform designed for corporate teams. It is no longer just a tool for assisting with writing and grammar as in earlier versions; rather, it integrates functions such as chatting, enterprise knowledge retrieval, repeatable workflows, AI agent development, model invocation, brand control, security governance, and usage auditing into a single platform.

The product serves markets, sales, customer support, human resources, operations, legal affairs, and technical teams; its goal is to enable organizations to transform one-time prompts into shareable, approvable, scheduled, and trackable business processes.

WRITER Agent

WRITER Agent is a unified interface for chatting and automation. Users can ask questions, upload files, create documents, analyze data, search the internet, produce presentations, and carry out multi-step tasks.

Unlike ordinary chatbots, it can invoke connectors, knowledge graphs, browsers, and coding tools, and carry out tasks in accordance with organizational rules.

Proxy outputs may still contain factual errors or deviations in tool performance. When it comes to external releases, customer communication, as well as changes to contracts and data, approval processes, access controls, and manual verification should be implemented.

Playbooks enable reusable workflows

Playbooks package prompts, input fields, knowledge, tools, and output formats into repeatable processes that teams can use to generate sales account summaries, analyze customer feedback, prepare RFPs, rewrite brand copy, or organize research reports.

Starter allows for the creation and execution of up to 5 Playbooks, which can only be shared within a team; Enterprise enables the creation of an unlimited number of Playbooks, Routines, and cross-team workflows, and offers more advanced capabilities for orchestration and approval.

Routines: Scheduled tasks

Routines allow Playbooks to be scheduled on a daily, weekly, or monthly basis, enabling the automatic execution of reporting, monitoring, and content preparation tasks. Starter versions support up to 3 scheduled workflows, while Enterprise versions offer an unlimited number of such workflows.

Timed processes should include mechanisms for notifying in case of failures, setting cost limits, and checking the timing of data updates. For reports that are generated automatically and rely on external connectors, it is also necessary to address issues such as expired permissions and changes in the structure of the source data.

Over 100 pre-set agents

The platform offers over 100 pre-set proxy directions that cover common departmental tasks; teams can use them as is or adjust them according to their own processes. Enterprise also allows administrators to manage the Agent Library in a centralized manner, thereby preventing employees from creating redundant or non-compliant processes arbitrarily.

Palmyra enterprise model

WRITER has developed its own series of large language models named Palmyra; the current developer documentation focuses on Palmyra X6, X5, and X4. These models support functions such as chatting, text generation, tool invocation, and enterprise workflows, and the platform also allows the integration of other large language models preferred by an organization within the Enterprise environment.

Model versions are updated and made obsolete. Production applications should use a fixed model ID, conduct regression tests, and keep an eye on the dates when models become obsolete; it cannot be assumed that the old Palmyra Med, Fin, or Vision endpoints will remain available indefinitely.

Knowledge Graph RAG

A Knowledge Graph transforms corporate documents and data into a relational knowledge structure, enabling agents to generate enhanced search results based on internal information. It is suitable for dealing with cross-document issues, complex data sets, answering questions related to policies and products, and showing the specific source segments that support the answers provided.

The RAG mapping approach can reduce responses that rely on information outside the company’s resources, but it cannot eliminate errors. Knowledge sources need to be continuously updated, deduplicated, and subject to access controls; moreover, answers should distinguish between facts from the original sources and conclusions derived by the model.

File and data connector

Starter offers basic connectors, up to 3 of which can be activated, along with a limited Knowledge Graph. Enterprise provides more comprehensive enterprise and industry data sources, an unlimited number of connectors, and more detailed control over knowledge access rights.

The connector brings the content from external systems into the AI context. Before deployment, permissions should be synchronized based on user identities, to prevent employees from accessing files to which they have no rights through proxies.

Brand tone and personality

Teams can set brand tone, style guidelines, and personality traits to ensure consistency in the output from different employees and workflows. The Starter version includes 1 team personality trait, while the Enterprise version allows multiple brand and voice profiles to be created for different departments.

Brand rules are suitable for regulating wording and format, but they cannot replace fact-checking, trademark regulations, and legal approvals. The rules themselves should also be versioned to prevent the reuse of outdated product information.

Canvas and multimodal output

The WRITER Agent allows for further editing of the results within Canvas, and it can generate various types of outputs in multiple formats, such as text, files, image analyses, image creations, presentations, and data analyses. Users can combine research, creation, and modification within a single task.

Images, charts, and presentation content should be checked for copyright issues, data labels, and accessibility. For automatically generated files, the layout and any sensitive information must also be verified before they are sent to the client.

Browser automation and manual takeover

The platform offers capabilities for browser automation and takeover, allowing agents to carry out various steps within web processes, with the user taking over control when necessary. It is suitable for data collection and repetitive data entry, but tasks such as logging in, making payments, deleting data, and submitting information externally are considered high-risk operations.

Companies should employ the principle of least privilege, sandbox environments, allowlists, and approval processes; they must not permit agents to execute tasks on unknown websites without restrictions.

AI Studio

Writer AI Studio is a full-stack development platform for building, deploying, and managing AI agents, offering options such as code-free agents, the Python Agent Builder, as well as APIs and SDKs. The Starter version provides limited access, while the Enterprise version offers full functionality, deployment options, hosting services, and higher API limits.

The development team can gradually transition from prototypes to production services, while managing models, knowledge, tools, security policies, and monitoring data through the same control interface.

Code-free Agent

Business users can create text-generation and research-oriented no-code agents; once deployed, these agents can be invoked through the Applications API, and used as tools for internal microservices or other agents. The old API name, application, is still used to maintain compatibility.

Agent Builder

Agent Builder is designed for Python developers; it enables the orchestration of tools, states, knowledge, and user interactions, and is suitable for complex agents that require custom logic. Applications can be integrated with Palmyra, third-party models, Knowledge Graphs, and enterprise systems.

Proxy development should include test data, validation of tool parameters, handling of timeouts, retry mechanisms, idempotency, and audit functions; it cannot be deployed simply by using demonstration prompts.

API and official SDK

WRITER offers REST APIs as well as official Python and Node SDKs, which enable access to functions such as chatting, text generation, streaming responses, tools, Knowledge Graph, and codeless proxies. API keys should be stored in secure environment configurations and must not be included in code repositories.

The official SDK reduces the costs associated with integration, but API calls still rely on hosted services and allocated usage quotas. Rate limits, model availability, and data regions are determined by the organization’s control panel.

Tool invocation

The Palmyra model allows function parameters to be selected and entered based on JSON structures, enabling serial or parallel invocation of enterprise tools. Developers can also use the already-deployed codeless agents as tools, so that higher-level agents can reuse specialized processes.

The model suggests that tool invocation does not guarantee parameter security; the server must re-verify the identity, scope, and business rules, and require explicit confirmation for irreversible actions.

Model and system interoperability

Enterprise supports the large models and systems preferred by organizations; the WRITER directory can also host certain third-party models. This allows teams to choose different models based on their specific tasks, while still maintaining unified governance and cost transparency.

Different models have varying data policies, capabilities, and prices. Before enabling them, administrators should review the provider’s terms and determine through testing which tasks can make use of external models.

Guardrails and governance

The platform offers content security, personal information protection, organizational policies, third-party Guardrails, and role permissions. Enterprise versions provide more detailed governance options, better observability, audit capabilities, customizable roles, SAML SSO, SCIM, and cross-team policies.

Guardrails represent a layer of controls designed to reduce risks, but they are not absolute barriers. Companies still need data classification, incident response mechanisms, red team testing, and manual approval processes.

Observability and auditing

Administrators can view AI Studio requests, responses, delays, and Token usage, and track costs by user, team, and Playbook. For sensitive environments, it is possible to configure the option to hide the request and response contents from the observation logs.

Logs can help in identifying errors and determining the causes of costs, but they may also contain sensitive data. It is necessary to set retention periods, define access roles, and establish rules for exporting logs.

Data privacy and security

The official package page states that customers retain ownership of their data; the platform follows a policy of zero data retention, and customer data is not used to train or improve models. WRITER offers transmission and static encryption, as well as compliance with regulations such as GDPR, CCPA, HIPAA, SOC 2 Type II, and PCI standards.

The specific certification reports, BAA agreements, and enterprise terms need to be determined in accordance with the plan. If third-party models or external connectors are used, it is also necessary to evaluate the data processing methods employed by those suppliers.

Starter plan

Starter is designed for fast teams that are just beginning to explore AI; it supports up to 5 users and includes a WRITER Agent, up to 5 Playbooks, 3 Routines, 1 team Personality, basic connectors, a limited Knowledge Graph, basic Guardrails, as well as single-team management.

The official website offers a 14-day free trial, without the need for a credit card. The package details page only indicates whether payment is monthly or annual, based on the number of seats and includes a fixed amount of credits; it does not show a unified price per seat in dollars.

The actual price and quota shall be based on the purchase page after registration.

Enterprise plan

Enterprise uses a sales inquiry system; it supports an unlimited number of users, teams, and free collaborators, without any restrictions regarding workflows or connectors. It offers a full-featured AI Studio, advanced orchestration and approval functions, cross-team sharing, interoperability between enterprise models, fine-grained governance, as well as priority support.

At the time of purchase, it is necessary to clarify the costs related to Pro and Lite seats, credits, Knowledge Graph capacity, connectors, models, support services, data areas, as well as the minimum contractual commitment.

Non-profit and educational discounts

Currently, the authorities offer a 20% discount on the Starter and Enterprise versions for non-profit organizations and educational institutions; an application must be submitted using an email address belonging to such an eligible organization. Eligibility and the final price are subject to approval by the sales team.

Price and version comparison

Palmyra API prices

The developer API is billed on a per-token basis. For Palmyra X6, the cost is 2 dollars for input and 8 dollars per million tokens for output;

The Palmyra X5 costs $0.60 to operate and generates $6 in revenue; the Palmyra X4 costs $2.50 to operate and generates $10 in revenue.

The WRITER Agent is charged based on the rate of the selected model, with 1 credit equivalent to $1 in terms of billing value. The prices of third-party hosted models can change, and these prices will be displayed in AI Studio before they are activated.

Price of Knowledge Graph

The current developer rates are 0.085 USD per GB per day for storage, 0.00015 USD per page for data extraction, 0.055 USD per page for OCR file processing, and 0.12 USD per page for Web Access. The Enterprise Data Connector is available under an Enterprise contract.

When dealing with a large number of scanned PDFs, the costs associated with OCR and storage can be significantly higher than those related to model tokens. The budget should take into account the expenses for parsing, hosting, retrieval, as well as the usage of models and connectors.

Tools API pricing

The cost of PDF Parser is currently $0.055 per page. The total cost is influenced by the number of pages in complex documents, the OCR requirements, and any need for repeated processing; applications should cache the results of processed tasks and set limits on file size.

Cancellation and refund

Users can cancel this option in the billing settings of their organization; the payment feature remains active until the end of the current cycle, after which the data associated with it may no longer be accessible. Refunds are determined on a case-by-case basis by WRITER, and the company does not guarantee that refunds will be provided.

GitHub and the open-source status

The WRITER full enterprise platform, the Palmyra model, Knowledge Graph, and hosted Agent services are not open-source products. The official GitHub repository provides Python and Node SDKs under the Apache 2.0 license, as well as an open-source Writer Framework, tutorials, and some agent examples.

The fact that the SDKs and frameworks are open source does not mean that the Palmyra weights or the enterprise backend can be made private. To use the open-source clients, API keys and hosting services are still required, and the deployment boundaries should be determined based on each individual component.

WRITER User Guide

Complete a basic task.

1. Clarify the issue, time frame, location, source priority, and output format;
2. In WRITER, upload the materials for which you have permission to use, or enter a search query;
3. First, use the WRITER Agent to establish a framework, and then employ Playbooks to add evidence through reusable workflows.
4. It is necessary to distinguish between factual information from the source, the author’s opinions, and AI-generated conclusions.
5. Check each item for dates, numbers, the original location, and any conflicting evidence;
6. The conclusions are manually revised, the verification time is recorded, and then they are published;

Create reusable professional workflows

1. Break down complex topics into four categories of questions: background, data, comparison, and conclusions;
2. By combining the WRITER Agent with Playbooks – which enable the reuse of workflows – and scheduled tasks via Routines, fixed research steps can be established.
3. Give priority to using the official website, research papers, regulatory documents, and raw data;
4. A second person is assigned to review conclusions that are considered high-risk;
5. Save queries, evidence, versions, and unresolved issues;
6. Re-run after the data changes and update the conclusions;

Which users is it suitable for?

• Organizations that need to deploy enterprise AI in a unified manner, rather than using separate chat tools;
• Teams that need to provide answers based on internal knowledge and cite their sources;
• Departments that wish to convert these tips into reusable Playbooks and scheduled workflows;
• Companies with requirements regarding branding, privacy, approval processes, and auditing;
• Teams that develop AI applications using APIs, SDKs, or Agent Builder;
• Technical leaders who are capable of orchestrating tasks across different models, connectors, and business systems are needed.

Product advantages

• Agents, knowledge, models, development, and governance together form a complete platform.
• Knowledge Graph is suitable for complex corporate data and multi-hop problems;
• Playbooks and Routines enable processes to be repeated and run automatically;
• It provides interoperability between its own Palmyra model and third-party models;
• It offers comprehensive enterprise permissions, auditing, and security features.
• Official SDKs and open-source frameworks facilitate development and integration.

Restrictions and Precautions

• WRITER is more suitable for organization-level deployment; when an individual only wants to write short articles, the configuration and associated costs may be high.
• Starter includes limitations on users, Playbooks, Routines, connectors, and KnowledgeGraph; full functionality is available only in Enterprise versions, for which a quote is required.
• Charging is based on agents, credits, tokens, the number of document pages, and storage, resulting in a complex budgeting structure;
• Models, agents, and knowledge bases can all make mistakes; business audits should not be skipped just because enterprise governance functions are available.

Frequently Asked Questions

Is WRITER still an AI writing tool?

Writing remains one of the capabilities, but the core focus at present is on enterprise Agentic AI platforms, which encompass knowledge bases, workflows, agent development, model APIs, and governance.

Can WRITER be used for free?

Starter offers a 14-day trial period without the need for a credit card. The official website does not disclose any permanent, completely free solution, nor does it provide information on the uniform price in dollars for the Starter service.

How much is the WRITER API?

For Palmyra X6, the cost for input/output is 2/8 dollars per million tokens; for X5 it is 0.6/6 dollars, and for X4 it is 2.5/10 dollars. The processing of knowledge graphs and PDF files is charged separately on a per-use basis.

Will customer data be used for training?

The authorities state that customers retain ownership of their data, and by default such data is not used to train or improve models; when third-party models and connectors are utilized, it is still necessary to check the policies of those suppliers.

Is WRITER open source?

The complete platform and models are not open source, but the official client and development components such as Python, Node SDK, and Writer Framework are open source.