Writer

WRITER Agent writes into systems Writer does not own. The Asana connector lets it "access available workspaces and team members, create, read, update, and delete tasks and subtasks, and retrieve task details, comments, dependencies, and project structure". The Salesforce integration is built by Writer, works in both directions, is published on AgentExchange, and lets the agent "query CRM data, manage records, run reports, and automate workflows".

Connectors has its own place in the main product navigation, with Microsoft SharePoint, Semrush, Snowflake, HubSpot and Microsoft Teams in the in product picker alongside Asana and Salesforce. Worked examples write across systems, from "Update the GTM report in SharePoint with campaign metrics" to a Snowflake query feeding a campaign that ends with emails actually delivered.

Connector Profiles let administrators configure several versions of the same connector with different permissions per team, and every agent request to a connected system passes an identity and permission check. Custom connections are open too, since agents can discover and call external tools through MCP connectors, one of the nine capabilities Writer scores.

Sourcewriter.com/product/writer-agent connectors, writer.com/product/connectors, support.writer.com what's new Asana and Salesforce entriesread 2026-09-13

Given a goal, WRITER Agent "autonomously plans and executes work across your data and tools", and once told what to accomplish, "it builds the plan to get there, mapping every step across the right data and tools".

A worked example chains a whole job, analyzing churned customer segment data in Snowflake, selecting top customers per segment, mapping product SKUs to segments, creating personalized email content, then deploying test emails and verifying delivery, tracked as a three section checklist with subtasks completing in sequence.

On duration, Writer says "X6 holds one goal for up to eight hours", planning, executing, testing its own output, correcting and delivering finished work, and a sample run timeline shows Plan, Execute, Optimize and Deliver phases, with events at 01:00 "Plan locked: 41 tasks", 03:42 "Self-test failed then fixed", and 08:00 "Shipped: verified", so the agent backtracks and recovers from errors inside a single objective.

Agents can delegate work to spawned sub-agents and merge the results, one of nine capabilities Writer scores, and AI Studio composes ready-to-use agents as building blocks that "interact with other ready-to-use and custom agents to automate complex, end-to-end business processes". The durable unit is the Playbook ("repeatable workflows, not one-off prompts"), with version control, edit history and restore in AI Studio.

Sourcewriter.com/product/writer-agent how-it-works, writer.com/models endurance sectionread 2026-09-13

The Knowledge Graph, now sold as Enterprise Brain, is a persistent graph that connects agents to enterprise sources for retrieval. Grounding, which Writer defines as "knowledge-graph and document grounding, retrieval, and citation", is one of nine capabilities it measures and publishes a score for. Answers cite their sources, alongside activity traces and decision logs, and Writer's evaluation describes pulling twenty-two sources through a knowledge graph and citing each one.

Connectors including Microsoft SharePoint, Snowflake, HubSpot, Salesforce, Microsoft Teams and Semrush feed the graph, with direct document upload alongside, and ready-to-use agents are "grounded in business context via WRITER's Knowledge Graph and document upload". Web search is a separate capability that admins switch on or off, using a standard search engine for current outside information, while grounding rests on the customer's own indexed content.

Sourcewriter.com/models capability table and FAQ, writer.com/product/enterprise-brain nav, writer.com/trust explainabilityread 2026-09-13

During a run, WRITER Agent "knows when to act and when to ask", and it "autonomously executes the plan and stops along the way for input", so the user stays "in full control without micromanaging every step".

Approval before deployment is a separate layer, since "governance is built into the agent lifecycle, from pre-deployment approvals to monitoring in production", and guardrails, identity validation, granular permissions and approval flows shape how agents interact with connected systems. Admins set limits ahead of time.

In AI Studio, Capabilities settings enable or disable high level functions per organization, browser automation is scoped to an allowlist or blocklist of named domains, web search is toggled separately, and agents are confined to the connectors, tools and knowledge sources they need.

Configurable guardrails "detect and redact PII and screen outputs against copyrighted material", "before a single word leaves the building", with prompt injection defense on every connection between an agent and a system. Under "Govern workflows at scale", an admin can reassign ownership of playbooks and skills or disable them entirely.

Sourcewriter.com/product/writer-agent, writer.com/trust safety and governance, support.writer.com Manage WRITER Agentread 2026-09-13

On certification, Writer says "We undergo annual SOC 2 Type II evaluations and hold ISO/IEC 27001, 27701, and 42001 certifications, confirming our implemented controls for security, privacy, and responsible AI management." It also covers HIPAA, GDPR and PCI-DSS, and ISO 42001 certifies an AI management system.

On access, Writer says "We support all major SSO providers and multi-factor authentication; role-based permissions and least-privilege access helps ensure every action is appropriately authorized." SSO and SCIM, granular RBAC, audit logs and customer managed encryption keys cover automated provisioning, a role model, federation and keys the customer holds.

For agents, "agent requests to connected systems pass through a governed execution layer that validates identity, checks permissions, and records activity", with scoping per connector so users and agents "only have access to the connectors, tools, and knowledge sources they need".

Identity is validated on every agent request, not only at login, and tenant isolation at the infrastructure level and data separation between teams within one org sit alongside.

Sourcewriter.com/trust security and transparency sections, writer.com/models enterprise-grade sectionread 2026-09-13

Every agent action can be audited, in Writer's words "Audit every agent action. Get complete logs of every user action and agent session, searchable and exportable to your SIEM, and retained for compliance." Admins can enable WRITER Agent session logs and review user feedback in AI Studio, with session records and thumbs up or thumbs down feedback viewable in the product and exportable as JSONL or CSV for offline review.

Reporting and logs cover usage, performance and costs in real time, token consumption is tracked live with alerts and hard blocks above target limits, adoption trends filter by team, billing group or seat, and each playbook has an Analytics tab showing runs, active users and submitted feedback. For explainability, "activity traces, source citation, and decision logs provide clarity into how AI reached an answer", and activity traces and performance reports help teams "monitor AI behavior, understand request patterns, and review how workflows are executed".

Sourcewriter.com/product/ai-studio how-it-works section, writer.com/trust transparency and control section, support.writer.com what's newread 2026-09-13

Agent memory that carries across sessions has arrived, introduced alongside Palmyra X6 as "Agent Memory to retain context across sessions", so state carries between runs. Writer does not say whether memory is scoped per user separately from the workspace, how long it lasts, whether it can be expired or purged on its own rather than with the account, or how it is read and written. Users can also "customize WRITER Agent's personality with different voices and custom instructions", plus reusable voice profiles.

Sourcesupport.writer.com what's new Agent Memory entry 2026-08-06, readread 2026-09-13

Self hosting is on offer, in Writer's words "Self-hosting and full data residency are available", alongside "managed, multi-cloud options, including dedicated private cloud deployment". Multi cloud is a choice among managed options, and the bring your own model path reaches into AWS Bedrock, Microsoft Azure and NVIDIA NIM on the customer's side, so the buyer picks where inference runs as well as where data sits.

Data controls sit alongside, with customer managed encryption keys, tenant isolation at the infrastructure level, zero retention by default, and a configurable organization wide automated data deletion schedule. Writer does not publish a list of regions, so the residency jurisdictions on offer are not stated.

Sourcewriter.com/models enterprise-grade section and writer.com/trust privacy and compliance sectionread 2026-09-13

The agent library at writer.com/agents/ is, in Writer's words, "a library of over 100 ready-to-use agents". They include the chart analysis agent, a "multimodal agent powered by Palmyra Vision" built to extract insights and perform contextual data interpretation from complex charts, the company financials research agent and the report generator agent.

Ready-to-use agents are "quickly grounded in business context via WRITER's Knowledge Graph and document upload" and serve as "composable building blocks in Agent Builder", interacting with each other to automate end to end processes. Playbooks and Skills add two more packaged layers, each with its own gallery.

Playbooks, billed as "repeatable workflows, not one-off prompts", come as structured units with an Overview, a "What it does" statement and a list of "Deliverables you'll get", and one example, an event outbound materials generator, returns email templates, LinkedIn messages, a cold call script and a materials guide.

Skills are browsable from a grid with named entries including SEC Filing Parser, Finserv Compliance Checker, Weekly Campaign Performance and Voice of Customer Summarizer.

Sourcewriter.com/blog/writer-agent-press-release 100+ ready-to-use agents, writer.com/blog/writer-ai-hq-press-release named agents, writer.com/product/writer-agent Playbooks and Skillsread 2026-09-13

Agent work starts on a clock, on external events and from meetings, not only when a person asks. Schedule based triggers, first launched as Routines to "schedule playbooks to run on a daily, weekly, or monthly basis, so your reports are ready when you are" and since renamed Triggers, run playbooks automatically on a recurring schedule, and event based triggers sit alongside them. Meetings are a third source.

WRITER Agent "joins the call, transcribes in real time, and gives every attendee the same summary of decisions and next steps", then kicks off the follow up work, and the meeting surface offers "Run a playbook from the meeting". Channels include the web application, WRITER for Slack, a Salesforce integration that works in both directions and is published on AgentExchange, the meeting surface, and the developer API at dev.writer.com.

Sourcesupport.writer.com what's new schedule-based and event-based triggers, writer.com/product/writer-agent meeting sectionread 2026-09-13

Palmyra X6 is the default model but not the only one. In Writer's words, "X6 leads our evaluation, so it's the default, but it isn't the only option. Admins can enable models from other providers, and users can choose a model at the start of a session. Specialized models can be selected for specific capabilities, such as image generation. And beyond WRITER's own catalog, admins can bring their own models from AWS Bedrock, Microsoft Azure, and NVIDIA NIM."

Admins enable models per provider in a control panel headed "Set what's available: control what models your teams can access", with a card for each provider (WRITER 8 of 8, Anthropic 9 of 9 and Google 0 of 5 models enabled) and each model toggled individually. "Users can pick the model they'd like to use for each session" from a searchable selector naming Palmyra X6, Palmyra X5, Claude Sonnet 4.6 and others, with the price of each model shown.

A separate preset governs specialized models, with GPT-5 and Gemini 3 Pro Image Preview for image generation. Writer also publishes its own comparison chart against rival frontier models, and its positioning argues against "stitching together your own stack of third-party models", though the controls above let customers use other providers.

Sourcewriter.com/models multi-model support section and FAQread 2026-09-13

A developer site at dev.writer.com carries an introduction at dev.writer.com/home/introduction and a versioned model reference at dev.writer.com/home/models#palmyra-x6. Outside applications can call Writer in two ways.

The Palmyra models are distributed through Amazon Bedrock, so a customer's own application can call them outside the WRITER application entirely, and the Salesforce integration works in both directions, so Salesforce can drive Writer as well as being driven by it.

AI Studio "fits right into your existing tech stack", integrating with the customer's monitoring tools, business systems, models and AI guardrails, and the Datadog Logs Plugin forwards LLM requests and responses to Datadog. MCP tool use runs outward from Writer to other tools.

Sourcewriter.com/product/ai-studio and writer.com/models Explore docs links to dev.writer.com, footers across the estateread 2026-09-13

Live metrics cover agents already in production, but Writer does not describe a way for a customer to test a change before it ships. Each playbook has an Analytics tab showing runs, active users and user feedback, thumbs up and thumbs down are captured on each agent output, viewable in AI Studio and exportable as JSONL or CSV, real time reporting covers usage, performance and cost, and an adoption view filters by team, billing group or seat.

These all read on what is already running. Playbook version history lets an owner "track edits, view version history, and restore previous versions", so a change can be rolled back, though not tested first.

Writer's nine dimension evaluation, "graded 0 to 1 against a locked baseline", measures Writer's own model rather than a customer's agents, the agent's "self-test failed then fixed" step is self correction inside a single objective, and approvals before deployment decide whether a person permits a launch rather than measuring how the agent performs.

Writer does not describe a harness customers run, scored test cases, a regression suite for a playbook edit or an A/B comparison of two agent configurations.

Sourcewriter.com/product/ai-studio, writer.com/models evaluation FAQ, support.writer.com playbook analytics and version controlread 2026-09-13

WRITER Agent has browser automation, governed from the Capabilities tab in AI Studio. In Writer's words, "Browser automation allows WRITER Agent to open a web browser in its secure sandbox environment. With this capability, the agent can perform tasks such as navigating to websites, reading page content, clicking on elements, and filling out forms."

Under "Allow WRITER to use browser automation for", an org or IT admin picks "All websites", "Only specific websites" (marked Recommended) with an included URL list, or "Exclude specific websites" with a blocklist. Adding a domain automatically includes its subdomains, and wildcards are not supported. Writer does not say how the agent copes when page layouts change on sites it does not own. Web search is a separate capability with its own admin toggle that "enables WRITER Agent to use a standard search engine to find information and images from the web".

Sourcesupport.writer.com Manage WRITER Agent, with writer.com/product/ai-studio capability controlsread 2026-09-13