Enterprise AI agents with company Knowledge Graphs, reusable skills and playbooks, connected actions, approvals, and editable deliverables.
WRITER
Explore features, practical uses and pricing below.
WRITER is an enterprise AI platform for turning company knowledge and working practices into reusable agents and workflows. Its current scope includes research, analysis, content creation, connected-system actions, and deliverables such as presentations. WRITER Agent provides the conversational interface, while playbooks, skills, Knowledge Graphs, connectors, and administrative settings determine how recurring work should run.
That structure is useful when a team needs more than an individual writing assistant. A sales operations group might want a repeatable account brief built from approved information. A marketing team might need a presentation in a shared visual style, with a reviewer approving the message before distribution. WRITER brings those inputs and instructions together, but the quality of the workflow depends on how the team defines them.
The WRITER Agent guide describes a natural-language interface that can use files, selected Knowledge Graphs, playbooks, and tools. Its capabilities include web research, data analysis, file generation, browser work, and presentation creation. Available capabilities depend on the organization's configuration and plan.
A useful starting request names the output, audience, sources, and boundaries. Instead of “analyze the account,” ask for a one-page account brief using the attached opportunity notes and an approved internal source, with unanswered questions listed separately. That gives the agent a deliverable to build and gives the reviewer a concrete standard.
Keep the conversation around one project. When the first result is incomplete, ask for a specific change: add the missing requirement, replace an unsupported claim, or separate customer statements from internal interpretation. A focused revision is easier to assess than asking the agent to make the whole result better without explaining what was wrong.
Knowledge Graph organizes connected and uploaded information for retrieval in WRITER. Supported sources include services such as Confluence, SharePoint, Google Drive, and Notion, as well as websites and manually supplied files. Authorized administrators and builders configure the collection and its team availability; the full enterprise setup has its own access requirements.
The important decision is what belongs in a particular collection. A customer-support graph might contain current policy and approved product guidance. A product-marketing graph might contain positioning documents and release material. Mixing archived and current policy without a clear convention can make a plausible answer difficult to trust.
The usage guide explains selecting the graph for a request and opening source excerpts. Ask a question that requires evidence, such as which document defines an eligibility rule, rather than requesting only a polished conclusion. The cited passage helps you check whether the source supports the precise answer.
Website-based knowledge is governed by the addresses you add. WRITER's web connector guide distinguishes a single page from included subpages and explains that links outside the configured domain or directory are not automatically part of the collection. A page linking to a separate documentation site does not by itself make that second site searchable.
For a product knowledge collection, map the material people actually need: product overview, supported configuration instructions, and relevant policy pages. Check that a question about each of those topics reaches the intended source. If an important answer is missing, investigate collection scope and synchronization before rewriting the prompt repeatedly.
Assign ownership to the collection as well as the resulting workflow. A playbook can remain unchanged while its source policy changes. Someone needs to know which files should be replaced, which old pages should be removed, and whether the updated material is available before the next scheduled run.
Playbooks package repeatable work into steps with inputs, instructions, relevant context, and supported tools. They can be developed from a useful task and then published for reuse. A playbook is more specific than saving a general prompt: it describes how the work should proceed and what the result should contain.
For an account brief, the inputs could be a company name, a selected opportunity record, and a reporting period. The steps could retrieve permitted information, organize it into a brief, identify missing facts, and prepare an internal draft. Define what should happen when a required input is absent. A request to guess the missing detail makes the workflow harder to audit.
Try the playbook against more than the ideal example. Use an account with sparse records, one with conflicting notes, and one with a closed opportunity. Those cases reveal whether the instructions handle the exceptions that colleagues will encounter. Publish the procedure after the team agrees on the output and decision points.
Skills are reusable packages that supply specialized instructions, reference material, or execution resources. You can create custom skills through a WRITER Agent session and add them to the library. A skill can be reused across tasks rather than embedding the same detailed rules in every playbook.
A reporting skill might define how to label assumptions, structure an executive brief, or distinguish a customer quote from commentary. A presentation skill might define slide types and rules for displaying data. The playbook then decides which information to gather and when to apply that skill.
Make the skill narrow enough to inspect. A single package that attempts to cover all departments' reporting standards can contain contradictory instructions. A shared brief format with clearly defined sections is easier to maintain. When a standard changes, test the skill against a representative output before using it in a wider scheduled process.
Playbooks support approval steps that pause a run for a designated reviewer. The configuration can ask the agent to revise and resubmit after feedback or leave the next decision to the person running the playbook. Approval changes take effect through the published configuration. A scheduled workflow also pauses when it reaches a required approval.
Place the checkpoint where the reviewer has enough information to make the decision. For an external message, show the proposed text, recipient context, and supporting evidence. Asking for approval before the draft exists is less useful than asking whether a concrete version is ready to send.
The reviewer should know what approval means. A manager checking commercial terms has a different responsibility from an editor checking tone. Write the question accordingly and identify a substitute when the primary reviewer is away. Otherwise, a recurring workflow can quietly become a queue of unfinished runs.
WRITER's connector offering connects agents to business systems for reading information and taking supported actions. It describes granular permissions and action logging. The specific tools, authentication, and allowed operations determine what an agent can do in a connected account.
The connector usage guide distinguishes authorized personal connections from custom setups and explains selecting available tools and relevant objects. Reading a record, updating a field, and sending a message are different operations. Configure the operations needed for the playbook rather than assuming that a connection should expose everything.
For a sales handoff, retrieve the relevant opportunity details first and prepare the proposed update. Check the record identifier and field mapping before enabling the write operation. A correctly worded summary attached to the wrong account is still an incorrect business action. The same principle applies to selecting a channel, folder, or recipient.
Schedule-based triggers run playbooks at defined intervals and create a new Agent session for each run. The configuration includes timing, timezone, and required variable values. Triggers can be disabled and re-enabled without discarding their configuration. Runs requiring input or approval wait for that input.
A weekly pipeline brief should specify the reporting cutoff and timezone, not merely “Monday morning.” Decide whether a late source update should appear in the next report or prompt a rerun. The schedule starts the process; it does not establish the business meaning of the period being reported.
Event triggers respond to supported connected-system events, including selected email, channel, file, calendar, or completed-call events. Configure the relevant scope so that unrelated activity does not start the playbook. Review the resulting sessions to confirm that the events you intended are the ones that actually launch work.
WRITER provides personal, team, and organization-level voice profiles with different visibility. A personal voice stays private to its creator. Shared voices let a team apply a common writing approach. Representative examples and reviewed profile characteristics help define the intended output style.
Its style and terminology guide explains that those settings need to be connected to a voice for automatic application in generated Agent outputs. Administrators can set or restrict the available voice through a profile. Merely maintaining a style guide elsewhere in the account does not establish that every generated draft uses it.
Use examples from the type of content the voice will produce. A customer email and an annual report may both be on brand while following different conventions. Check terminology in the output as well as tone, particularly when a product name or approved phrase has a precise meaning.
Presentation generation can use a prompt, source material, a theme, and optional skills or playbooks. WRITER documents editable PowerPoint output with native elements such as tables and charts. That gives a reviewer a file to continue working on rather than only a flat image of a slide.
Themes can be personal or administered for the organization. They define visual foundations such as fonts, colors, and layouts. Local font availability can affect how a downloaded presentation renders, so open the final file in the application the audience will use.
For a recurring business review, define the message of each slide type before generating the deck: current result, comparison, explanation, and decision needed. Then review the actual chart labels and data. A theme helps with appearance; it does not decide whether a particular chart tells the right story.
Agent profiles configure capabilities, knowledge selections, instructions, voice, and sharing controls for a particular experience. Profiles can restrict or customize features such as web search and browser work. This lets an organization provide different setups for departments with different needs.
Administrators can also curate the team playbook library so colleagues can recognize the organization's selected workflows. Give each playbook an understandable name and purpose. “Weekly account handoff” is more useful to a new colleague than an internal experiment number.
Reporting and governance provide inventory and usage information, including workflow state, ownership, dependencies, and run activity within the documented scope. Use that information to find unused workflows, investigate failures, and understand which connection a process relies on. A successful run count alone does not reveal whether colleagues accepted the output.
The current plan page describes paid Starter seats with credit limits and an Enterprise offering arranged through sales. Starter has a time-limited trial, a small-team scope, a limited number of playbooks, and basic connectors. Enterprise adds broader orchestration, organizational controls, knowledge capabilities, and cross-team use. The page distinguishes seats that build and customize workflows from more limited enterprise access.
Request a configuration that matches the workflow you intend to operate: users who build, users who run, required connectors, knowledge storage, approvals, and expected usage. Confirm current credit allowances and any additional services in the commercial terms. A trial is useful for evaluating a defined process, but it is not evidence that every enterprise control is included in a smaller plan.
Its current platform covers broader enterprise knowledge work, connected actions, analysis, and generated deliverables. Marketing is one application; the same building blocks can support sales, operations, and other departments when configured for their sources and processes.
A skill captures a reusable way to perform part of a task. A playbook defines the sequence, inputs, and tools for a complete workflow. A presentation playbook, for example, can apply a shared formatting skill after gathering the required information.
Yes. A required approval pauses the run. Plan who will respond and what they need to check so that the workflow can continue when the review is complete.
No. Source excerpts make review easier, but you should still check that the selected passage supports the statement and that the collection contains the current material needed for the task.