Documentation

Getting Started

1. Sign in at workweaver.ai/auth using Google, Microsoft, or email. (Currently limited to invited organizations.)
2. Pick a workflow to start with: Morning Briefing, Inbox Triage, Lead Follow-up, or create your own.
3. Connect your tools in the Connection Center. Start with just what your chosen workflow needs (e.g., Gmail + Google Calendar for a morning briefing).
4. Review the plan. Workweaver shows you what it intends to do before it does it. Approve any gated actions.
5. Check the results. Use Mission Calendar for the timeline view, then inspect decision traces, proof cards, and the evidence trail.

Best Practices for Working with AI

Like any new team member, AI works best when you set it up well. Here are the practices that make the biggest difference.

1. Give clear background (context)

Share your goals, your customers, your brand voice, and any constraints. The more background Workweaver has, the better its decisions will be. Upload relevant documents, provide descriptions of your ideal customer, and describe what good outcomes look like.

2. Define boundaries (dos and don'ts)

Be explicit about what's allowed and what's off-limits. Which tools can it use? What types of messages should it send? What topics should it avoid? These boundaries are stored as rules in WorkMemory and enforced on every action.

3. Set clear acceptance criteria

For each workflow, define what "done" looks like. Is it a booked meeting? A CRM update? A summary email? Clear acceptance criteria help Workweaver focus on the right outcome and let you evaluate whether it's working.

4. One task at a time

Start with a single workflow. Get it producing consistent, good results before adding another. Spreading across multiple workflows too early dilutes the learning that makes each one better.

5. Review decision traces, then course correct

Check why Workweaver made each decision. If the reasoning is off, adjust the background or boundaries. If it's spot-on, let it keep going. Every correction compounds into better future decisions.

6. Understand before scaling up

Run workflows at low volume first. Review the outputs. Once you're confident in the quality, increase volume and loosen approval gates gradually. This mirrors how you'd ramp up any new team member.

Common mistakes to avoid

• Going fully autonomous before you've reviewed enough decision traces
• Providing vague goals ("do outreach") instead of specific ones ("reach VP-level contacts at SaaS companies with 50-200 employees about our security product")
• Connecting all tools at once instead of starting with the minimum set
• Ignoring approval requests instead of using them to steer
• Expecting perfect results from Day 1 without providing background or corrections

Workflows

Workflows are the core of Workweaver. They define a sequence of steps that run automatically, with approval gates at the points you choose.

Built-in templates

Start from templates designed for common patterns: sales outreach sequences, meeting follow-ups, inbox triage, research briefs, onboarding checklists, and more. Each template is a starting point you can customize.

Custom outcome loops (Flows)

In Flows, you design workflows as trigger-branch-action graphs. Define what triggers the workflow, add branching logic (e.g., "if replied, do X; if no reply after 3 days, do Y"), and chain actions together. Save, schedule, and monitor from the same interface.

Multi-step execution

Workweaver handles multi-step workflows end to end: research a prospect, draft an email, wait for approval, send, monitor for replies, follow up on schedule, update CRM, and log everything. Each step has its own decision trace.

Quick Agent

For one-off tasks, Quick Agent selects the right execution mode (solo, assisted, or team) based on the complexity and risk of what you're asking. It checks which tools are connected, estimates what's needed, and starts with the smallest safe configuration.

Decision Traces

Key actions Workweaver takes produce a decision trace. This is the record of what happened and why. Trace coverage is expanding toward full action tracing.

What's in a trace

• What was done: The specific action taken (email sent, CRM updated, meeting scheduled).
• Why: The reasoning behind the choice. What signal triggered it, what data informed it.
• What alternatives were considered: Other options that were evaluated and why they were not chosen.
• Confidence level: How certain the system was about this being the right action (0.0 to 1.0).
• What was learned: Any pattern or fact extracted for future decisions.

Why this matters

In traditional work, decisions happen in someone's head. There's rarely a record of why a particular prospect was chosen, why a specific message angle was used, or why a meeting was scheduled at that time. Decision traces make every choice visible and reviewable. This is not about surveillance. It's about building a shared understanding of how work gets done, so you can improve it.

WorkMemory

WorkMemory is how Workweaver builds institutional knowledge that persists and compounds. Unlike chat history or logs, WorkMemory is structured, searchable, and actively used in every decision.

Four types of memory

• Journal: A chronological record of what happened, auto-generated from execution evidence.
• Facts: Key-value pairs representing what's known ("Acme Corp prefers Tuesday morning meetings", "VP-level contacts respond better to ROI framing").
• Patterns: AI-discovered patterns with confidence scores ("Follow-up emails sent within 24 hours have 3x response rate for this ICP").
• Rules: Constraints you define ("Never contact prospects on weekends", "Always include a calendar link in meeting requests").

How it compounds

After every task, Workweaver extracts what worked and what didn't. Successful patterns are reinforced. Unsuccessful patterns decay. Over time, decisions get sharper because they draw on accumulated knowledge specific to your business, your customers, and your workflows.

Scope and privacy

Memory is private by default at the individual level. It can be promoted to team, department, or organization scope when appropriate. Cross-tenant learning never happens. What one organization's Workweaver learns is invisible to every other organization.

Approval Gates

Approval gates are how you control which actions require your sign-off and which can run automatically.

Default gates

By default, high-risk actions are gated: sending emails, WhatsApp messages, SMS, making phone calls, and deleting CRM records. These require your explicit approval before they execute.

Customization

You decide the pace of autonomy. As you review decision traces and build confidence, you can loosen gates on specific action types. You can also tighten them at any time. Emergency stop pauses everything instantly.

What approval looks like

An approval request shows: what Workweaver wants to do, why (with evidence citations), the risk level, and whether the action is reversible. You approve, edit, or reject with one action.

Connections & Integrations

Workweaver connects to the tools your team already uses. No data flows through any tool that isn't explicitly connected.

Direct integrations

Gmail, Outlook, Zoho (17 apps), Salesforce, HubSpot, Google Calendar, Outlook Calendar, Slack, WhatsApp, Twilio (voice and SMS), GitHub, Notion, Google Drive, Jira, Linear, and Stripe. Microsoft Teams and Zoho Cliq bridges remain guided-rollout integrations when tenant-specific provider configuration is complete.

Connector marketplace

Additional integrations are available through the connector marketplace and guided rollout. Connect via OAuth or API key.

Custom connectors

For tools not in the marketplace, you can build custom connectors using REST APIs, OpenAPI specs, webhooks, CLI tools, or the MCP protocol. If your tool has an API, Workweaver can connect to it.

Multi-channel workflows

Start a conversation on one channel (email) and continue it on another (WhatsApp, Slack, voice). Workweaver maintains conversation context across channels using a shared session model.

Security & Access Control

Role-based access (RBAC)

Three role levels: Admin (full control), Manager (team oversight), Agent (execution). Roles determine what each user can configure, approve, and access.

Capability-based access (CBAC)

Beyond roles, Workweaver uses capability-based access for AI execution. Each workflow gets a delegation warrant specifying exactly which tools, operations, memory scopes, and budgets are allowed. Capability leases enforce runtime limits: maximum concurrency, delegation depth, spend caps, and time bounds. This prevents scope creep during execution.

Tenant isolation

Every organization is fully isolated. Data, memory, evidence, and learned patterns never cross tenant boundaries. Enforced by middleware and verified by integration tests.

Evidence immutability

The Evidence Lake uses S3 Object Lock in COMPLIANCE mode with 7-year retention. Records cannot be altered or deleted by anyone, including Workweaver staff. Evidence is fully exportable at any time.

Data privacy

Your data is yours. Not even we can access it without your explicit permission. If you contact us for troubleshooting, we only access what you specifically authorize. Usage data for billing is separate from your business data. GDPR-compliant account deletion and data export are built in.

Evidence Lake

Key actions, decisions, and outcomes are recorded in the Evidence Lake. This is the immutable, long-term audit trail that makes Workweaver's work verifiable. Coverage is expanding toward full action tracing.

What gets recorded

• Every workflow execution with start time, duration, steps, and outcome
• Decision traces for key actions (reasoning, alternatives, confidence)
• Approval gate events (requested, approved, rejected, expired)
• Tool calls with request/response details
• Memory updates (facts learned, patterns discovered)
• Degradation events (when a provider was unavailable or fallback was used)

Retention and compliance

Evidence is stored in S3 with Object Lock in COMPLIANCE mode. 2,555 days (7 years) retention. KMS encryption at rest. Once written, evidence cannot be modified or deleted. This is verified via AWS API response and auditable at any time.

Export

Full evidence export is available in JSON and CSV formats. Your evidence trails are yours. No data held hostage.