Support Center | RunFrame

Start Here

Pick the path closest to what you are trying to do.

Admin

New client onboarding

Create the org, connect Google Drive, import employees, pay, provision, and invite the team.

Employee

Employee capture requests

Submit Loom walkthroughs and send examples, templates, screenshots, or checklists.

Founder

First interview workflow

Make sure the employee exists before the GHL booking and verify the Vapi call lifecycle.

Compliance

Security and PII review

Understand quarantine, redaction, approvals, and when to escalate sensitive samples.

Walkthroughs

These guides match the current production flow. Recordings use sanitized screens and avoid customer data.

Admin20-40 minutes

Onboard a new client

Use this for the first admin setup path in the RunFrame portal.

1Open portal.runframe.ai/login on desktop and create an account with a work email.
2Enter company details, team size, industry, and select Google Workspace.
3Authorize Google Drive. RunFrame uses the drive.file scope, so it only accesses files it creates.
4Import employees manually or with a CSV using name, email, title, role_category, and manager_email.
5Assign admin, manager, employee, and compliance reviewer roles.
6Review pricing, complete Stripe checkout, return to onboarding, and begin provisioning.
7After provisioning starts, send employee invitations from the invitation step.

Notes

Use a small first batch for the first live customer: 2 to 5 employees.
The onboarding wizard is intentionally desktop-only.
Teams over 100 employees should use founder-assisted onboarding.

Admin10 minutes

Prepare the employee CSV

Format employees so reporting lines and interview scheduling match cleanly.

1Create CSV columns: name, email, title, role_category, manager_email.
2Leave manager_email blank for at least one top-level employee.
3Make every manager_email match another employee email in the same CSV.
4Use work emails only. The first interview booking should use the same email as the employee row.
5Choose a supported role_category such as operations_manager, sales_rep, customer_success, administrative_ea, or other.
6Upload the CSV in the Org Chart step and fix any validation errors before continuing.

Notes

The current portal CSV import does not collect phone numbers.
Phone numbers for the first Vapi interview are collected through the scheduling flow.

Founder5-10 minutes

Schedule the first AI interview

Use this after the employee exists in RunFrame.

1Confirm the employee was imported in the RunFrame org chart.
2Send the RunFrame interview calendar link or manually create the GHL appointment.
3Use the same email address that exists on the RunFrame employee record.
4Collect the employee's direct phone number in the booking form.
5Book a 20 to 30 minute slot and allow up to 5 minutes for the scheduler to place the call.
6After booking, verify a scheduled phone skill request exists before the call time.

Notes

Do not schedule the GHL appointment before the employee exists in RunFrame.
Email matching is safer than phone matching for first-client onboarding.

Employee3-8 minutes

Submit requested artifacts

Use the capture page when RunFrame asks for Looms, examples, screenshots, templates, or checklists.

1Open portal.runframe.ai/capture while signed into your employee account.
2Review outstanding artifact requests, request IDs, due dates, and the prompt for each request.
3For a Loom, paste the Loom share URL into the request form and submit it.
4For files, screenshots, templates, checklists, or examples, reply to the request email or send them to artifacts@inbound.runframe.ai.
5Include the request ID in the email subject when possible.
6Return to the capture page to confirm recent submitted samples were received.

Notes

There is no file upload UI in the capture page yet.
Redact private customer details before sending examples when practical.

Founder15-30 minutes

Review generated skills and Cowork setup

Use this after the first employee interview and any requested artifacts are complete.

1Open the employee's Google Drive Skills folder created by RunFrame.
2Find the generated workflow package under skills/<workflow-name>/.
3Review SKILL.md for concise action instructions and trigger conditions.
4Review REFERENCE.md for transcript evidence, source details, and artifact state.
5Review CHECKLIST.md and EXAMPLES.md when present.
6If run-log.json and skills/weekly-review exist, confirm the workflow is repeating and judgment-based.
7Help the employee open Claude Desktop or Cowork against the provisioned workspace and start with start-of-day-sequence.

Notes

RunFrame does not install the customer's Claude app automatically.
First-client skills should receive operator review before first use.

ComplianceVaries

Review PII and quarantine decisions

Use this when a sample is flagged for sensitive information.

1Open the quarantine review queue when RunFrame notifies you that a sample needs review.
2Read the detected PII categories and redacted preview.
3Approve the redacted copy if it is safe to use for skill generation.
4Reject the sample if it should not be used.
5Escalate if you are unsure or the sample contains highly sensitive customer, borrower, financial, or regulated data.
6For early launch accounts, contact support if the self-service security screen does not expose the setting you need.

Notes

Original sensitive content is not sent to Claude for skill generation.
Tier 1 and Tier 2 originals remain quarantined until review or retention expiry.

Admin5-15 minutes

Read weekly reports

Use dashboard reports for employee, department, and executive visibility.

1Open Dashboard, then Reports.
2Select the latest executive, department, or employee report.
3Review risks, bottlenecks, repeated work, and recommended next actions.
4Use employee and department views to inspect local patterns before making broad changes.
5Ask support to investigate if a report is missing after the expected Friday run.

Claude & Cowork

Practical guidance for using Claude Chat, Claude Desktop/Cowork workspaces, Artifacts, Claude Design, Projects, Styles, connectors, file creation, and RunFrame-generated skill packages.

Choose the right Claude surface

Use Chat for thinking together, Projects for repeat context, Cowork for delegated desktop work, Chrome for browser tasks, and Design for visual exploration.

Use Claude Chat when you need a fast answer, rewrite, analysis pass, or one-off draft.
Use Claude Projects when Claude should remember the same knowledge base, policy, brand voice, or SOP across many chats.
Use Claude Desktop or Cowork when the job involves local files, folders, work products, and longer-running execution.
Use Claude in Chrome when the work happens across websites, forms, dashboards, inboxes, CRMs, or multiple tabs.
Use Claude Design when the output is a prototype, visual direction, deck, one-pager, or product experience concept.
Use RunFrame-generated skills when the work must follow a captured customer workflow instead of generic instructions.

Starter prompt

I need to complete [task]. Recommend whether I should use Claude Chat, a Project, Claude Desktop/Cowork, Claude in Chrome, Claude Design, or a RunFrame skill. Explain the tradeoff, required context, and review points.

Use Claude like a capable coworker

Give Claude the job, the context, the desired output, and the boundaries.

Start with the role: what should Claude act as for this task?
Explain the business goal and why the answer matters.
Attach or paste the relevant source material before asking for judgment.
Tell Claude the exact output format you want.
Name the constraints: privacy, approval requirements, tone, deadline, and systems of record.
Ask Claude to state assumptions and unresolved questions before taking action.
For agentic work, ask Claude to produce a plan before acting and to stop when permissions, data, or approval are missing.

Starter prompt

You are helping me complete [workflow]. Goal: [business outcome]. Context: [source details]. Use only the attached material and clearly label assumptions. Output format: 1) Summary, 2) Recommended action, 3) Risks, 4) Open questions, 5) Final draft.

Use Claude Chat for day-to-day work

Claude Chat is best when you want quick collaboration without giving Claude control of files, browser tabs, or external tools.

Open a fresh chat for each unrelated task so stale context does not leak into the answer.
Put source facts, goals, audience, and constraints in the first message.
Ask Claude to ask one or two clarifying questions only if the missing detail would change the result.
Use styles when you need Claude to be concise, formal, explanatory, or consistent with a team voice.
Ask for a critique pass before asking for a final rewrite.
Move repeatable work into a Project or skill once the same prompt is being reused.

Starter prompt

Help me with [task]. Audience: [audience]. Source material: [facts/files]. Constraints: [rules]. If anything essential is missing, ask up to two questions; otherwise produce the best first draft and flag what I should verify.

Run a RunFrame skill package in Claude Desktop or Cowork

Use the generated workspace files instead of starting from a blank chat.

Open the employee's RunFrame Drive workspace.
Read CLAUDE.md first so Claude understands the workspace routing rules.
Open the relevant skill folder under skills/<workflow-name>/.
Read SKILL.md for the task instructions and CHECKLIST.md for done-ness.
Use REFERENCE.md for source evidence and EXAMPLES.md only when it exists.
When the work is finished, summarize actions taken, skipped items, escalations, and final status.

Starter prompt

Use the skill at skills/[workflow-name]/SKILL.md. Read CHECKLIST.md and REFERENCE.md before acting. Do not invent missing details. If approval, access, source files, or decision criteria are missing, stop and ask.

Delegate a complete task to Claude Cowork

Cowork is strongest when the task has a clear deliverable, real source files, and human review points.

Name the finished deliverable: report, spreadsheet, cleaned folder, deck, memo, checklist, ticket, or draft.
Point Cowork at the folder, files, connected apps, or browser workflow it should use.
Attach the skill, SOP, checklist, examples, and quality bar before it starts.
Ask for a plan and approve the plan before it touches files or records.
Require a final summary with files changed, assumptions, skipped steps, and items needing human approval.
Keep destructive actions, external sends, payments, legal calls, and sensitive record changes behind explicit confirmation.

Starter prompt

Delegate this to Claude Cowork. Outcome: [deliverable]. Sources: [folders/apps/files]. Procedure: [skill or SOP]. Quality bar: [criteria]. Review rules: ask before changing records, deleting files, sending messages, or making external commitments. Return a plan first.

Use Claude in Chrome for browser workflows

Claude in Chrome can read, navigate, click, fill forms, schedule browser tasks, and hand browser context to Cowork when enabled.

Use the official Chrome extension and confirm your plan supports the beta.
Start with trusted websites and narrow site permissions while the team learns the workflow.
Use Ask before acting when the task involves forms, CRM records, invoices, customer accounts, or anything work-critical.
Record simple repeatable workflows only after you have tested them manually.
Pair Chrome with Cowork when the browser gathers data and Cowork should produce the finished workbook, memo, deck, or report.
Stop the run and escalate if Claude visits an unexpected site, asks for unnecessary data, or tries to take an action outside the approved plan.

Starter prompt

Use Claude in Chrome for this browser workflow. Goal: [outcome]. Sites: [trusted sites]. Actions allowed: [allowed actions]. Actions requiring confirmation: [sensitive actions]. First create a plan; after approval, proceed and summarize every record or page changed.

Connect Google Workspace and approved tools

Connectors let Claude retrieve context and sometimes take actions in tools such as Gmail, Calendar, Drive, Slack, Linear, Jira, and other approved services.

Confirm the connector is approved by your organization before authenticating.
Connect only accounts that match the work context; do not mix personal and company accounts.
Ask Claude to cite which emails, calendar events, docs, or records it used.
Limit the request to the minimum tool access needed for the task.
Review drafts and proposed changes before anything is sent, created, updated, or deleted.
Disconnect or revoke permissions when a connector is no longer needed.

Starter prompt

Use only approved connectors needed for this task. Retrieve the minimum necessary context, cite the records used, and ask before creating, updating, deleting, sending, scheduling, or changing permissions.

Prompt Claude for better business writing

Claude works best when audience, purpose, tone, source material, and review criteria are explicit.

Name the audience and channel: customer email, internal Slack, executive memo, SOP, or checklist.
Paste source facts separately from instructions.
Ask for a first draft and a short rationale.
Ask Claude to revise for a specific style such as concise, formal, explanatory, warm, or direct.
Before sending externally, ask Claude to flag unsupported claims and sensitive details.
Save high-performing instructions as a style, Project instruction, or RunFrame skill when the same writing pattern repeats.

Starter prompt

Draft a [channel] for [audience]. Goal: [goal]. Source facts: [facts]. Tone: [tone]. Constraints: [privacy/legal/approval constraints]. Return the draft plus a short list of claims I should verify before sending.

Use Claude Projects for repeat work

Projects are useful when Claude needs the same documents, SOPs, examples, or brand guidance repeatedly.

Create one Project per department, client, workflow family, or recurring initiative.
Upload durable materials: SOPs, brand voice, product docs, policy, examples, and approved templates.
Keep temporary drafts in chats, not in project knowledge.
When knowledge becomes large, rely on Claude's project retrieval instead of re-uploading files every time.
Refresh project knowledge when policy, pricing, templates, or role ownership changes.
Add Project instructions that tell Claude which sources are authoritative when files conflict.

Starter prompt

Use the project knowledge as the source of truth. If the project files conflict with this chat, tell me what conflicts and ask which source to use.

Create Artifacts and AI-powered mini-apps

Artifacts are best for standalone things you want to edit, reuse, preview, or share.

Ask Claude for an artifact when you want a document, diagram, webpage, dashboard, SVG, React component, or prototype.
Describe the user, job-to-be-done, fields, interactions, visual style, and constraints.
Iterate by pointing to specific sections or visible labels instead of asking for vague changes.
Use versions to compare changes and revert if a direction gets worse.
For AI-powered artifacts, remember that users authenticate with Claude and their usage counts against their Claude subscription.
For customer-facing artifacts, review the artifact for unsupported claims, sensitive data, mobile layout, and accessibility before publishing.

Starter prompt

Create an artifact for [use case]. It should help [user] do [job]. Include [fields/interactions]. Visual style: [style]. Make it easy to inspect, edit, and share. After creating it, list three improvements we could make next.

Use Claude Design for software experience exploration

Claude Design is an Anthropic Labs product for exploring software experiences, prototypes, decks, one-pagers, and exportable visual work.

Use it for early product screens, workflow concepts, visual directions, and experience options.
Start with the audience, task, screen type, and constraints.
Ask for multiple directions before committing to one.
Give feedback in product language: hierarchy, states, density, handoff, brand fit, and edge cases.
Use web capture or uploaded brand assets when the prototype should look like the real product.
Treat the output as a design exploration or handoff artifact, not production code without review.
Export or hand off the chosen direction into the team's normal design workflow.

Starter prompt

Explore three design directions for [software experience]. Audience: [audience]. Primary job: [job]. Required screens/states: [states]. Constraints: [brand/platform/accessibility]. Show tradeoffs and recommend one direction for handoff.

Use Claude file creation for reports and work products

Claude can create and edit files such as spreadsheets, presentations, documents, and PDFs on eligible plans.

Provide the source data and describe the final file type.
Specify structure: tabs, columns, formulas, charts, sections, or slide outline.
Ask Claude to explain assumptions and data quality issues.
Download and inspect the file before sharing externally.
Do not upload regulated or sensitive data unless your organization's Claude policy allows it.
For spreadsheets, ask Claude to include formulas, source notes, and a data quality tab when appropriate.

Starter prompt

Create a [spreadsheet/presentation/document] from this source material. Required structure: [structure]. Include formulas/charts/sections for [needs]. Before creating it, identify missing data or assumptions.

Use connectors and desktop extensions safely

Connectors let Claude work with approved cloud services; desktop extensions connect Claude Desktop to local apps and files.

Use remote connectors for cloud services your organization has approved.
Use desktop extensions when work should stay local to your computer.
Ask Claude only for data you are already allowed to access.
Review retrieved context before acting on it.
Avoid connecting personal accounts to company workflows.
For custom connectors, authenticate only trusted servers and review requested scopes before approving.

Starter prompt

Use only the connected sources needed for this request. Retrieve the minimum necessary information, cite what you used in plain language, and ask before drafting external messages or changing records.

Review Claude outputs before they reach customers

RunFrame workflows should make review explicit so Claude helps teams move faster without hiding judgment calls.

Ask for source citations or source labels whenever the answer depends on documents, emails, dashboards, or tickets.
Require Claude to separate facts, assumptions, recommendations, and open questions.
Have Claude flag privacy risks, unsupported claims, stale data, missing approvals, and customer-impacting changes.
Inspect any generated files, formulas, links, permissions, and recipients before sharing.
Turn repeated review failures into updates to the skill, checklist, Project instructions, or source templates.

Starter prompt

Review this output before I use it. Check factual support, source coverage, privacy risk, tone, missing edge cases, approval requirements, and whether it follows the provided skill/SOP. Give fixes first, then rewrite only if needed.

Manage usage limits

Long chats, large files, tool use, model choice, and artifact creation all affect usage.

Plan the question before sending it.
Batch related questions into one clear request.
Use Projects for documents you will reuse.
Start a new chat when the conversation has drifted or accumulated stale context.
Choose a lighter model for routine drafting and reserve the strongest model for high-value reasoning.

Starter prompt

Before answering, tell me if this request would be better handled as a shorter chat, a project, or separate tasks. Then proceed with the most efficient approach.

Reusable Prompt Patterns

Decision memo

Analyze [decision]. Context: [facts]. Options: [options]. Criteria: [criteria]. Return: recommendation, reasoning, risks, information still needed, and next action.

Workflow execution

Run this workflow using the attached instructions. Confirm inputs, follow steps in order, use documented decision rules, stop on missing access or approval, and return final status.

Cowork delegation

Outcome: [deliverable]. Sources: [files/apps]. Procedure: [skill/SOP]. Constraints: [rules]. Before acting, return a plan, files/tools needed, approval checkpoints, and what you will not do without confirmation.

Chrome workflow

Use Claude in Chrome on [trusted sites] to [goal]. Allowed actions: [actions]. Confirm before [sensitive actions]. Record sources, changed records, and anything that did not match the approved plan.

Connector research

Use approved connectors to answer [question]. Pull only the minimum context, cite the emails/docs/events/records used, separate facts from assumptions, and do not create or change records unless I approve.

Design exploration

Create three Claude Design directions for [experience]. Include audience, primary flow, states, brand constraints, density, accessibility notes, and a recommended handoff direction.

SOP creation

Convert this rough process into an SOP. Include trigger, inputs, tools, steps, decision points, escalation rules, checklist, examples, and owner. Mark assumptions clearly.

Artifact iteration

Update only [specific section/component] of the artifact. Preserve everything else unless a change is required. Explain what changed in 3 bullets.

Quality review

Review this output for correctness, unsupported claims, privacy risks, missing edge cases, tone, and whether it satisfies the stated goal. Give fixes before rewriting.

Prompt improvement

Improve this prompt so Claude has clear role, goal, context, source rules, output format, approval boundaries, and quality criteria. Explain the changes briefly, then provide the final prompt.

Official Anthropic References

Claude product features change quickly. These links are the source material behind this section.

Getting started with ClaudeClaude Help Center Prompting best practicesClaude Platform Docs Usage limit best practicesClaude Help Center What are artifacts and how do I use them?Claude Help Center AI-powered artifactsClaude tutorials Claude DesignAnthropic Claude for Creative WorkAnthropic Claude CoworkAnthropic Claude for ChromeClaude Get started with Claude in ChromeClaude Help Center Use connectors to extend ClaudeClaude Help Center Use Google Workspace connectorsClaude Help Center Custom connectors using remote MCPClaude Help Center Agent SkillsAnthropic Configuring and using stylesClaude Help Center What are Projects?Claude Help Center Retrieval Augmented Generation for ProjectsClaude Help Center Installing Claude DesktopClaude Help Center Create and edit files with ClaudeClaude Help Center

FAQs

Is RunFrame fully self-serve right now?

The production system is live, but first-client onboarding should still be founder-assisted. The portal, API, workers, capture flow, Drive provisioning, and reports are live; the first few customers should be monitored through setup and first interview.

Which cloud workspace is supported?

Google Workspace is the supported launch path. Microsoft 365 and other stacks are not live self-service options yet.

What Google permissions does RunFrame request?

RunFrame requests Google's drive.file scope. That lets RunFrame create and manage files it creates; it does not grant access to all existing Drive files.

Can I onboard from a phone?

No. The onboarding wizard is desktop-only because org chart import, billing, Google OAuth, and access review are too dense for mobile.

What email address should employees use for files?

Employees should reply to the artifact request email or send files to artifacts@inbound.runframe.ai. They should include the request ID when possible.

Where do I submit a Loom?

Sign in and open /capture. Paste the Loom URL into the matching request card. If no request is visible, use the general Loom submission form on the same page.

Why does /capture say my account is not linked to an employee profile?

Your login exists, but it has not been connected to an employee row in the organization. Ask your admin to resend the invitation or contact support.

Does Claude see unredacted sensitive samples?

No. Samples with high sensitivity are quarantined and redacted before skill generation. Original sensitive content remains in private storage for review.

Does RunFrame automatically install Claude Cowork?

No. RunFrame creates the Drive workspace, skill packages, and setup-ready events. A founder, operator, or customer admin still helps the employee connect Claude Desktop or Cowork during early launch.

Should I use Claude Chat, Claude Cowork, or Claude in Chrome?

Use Chat for quick collaboration, Projects for repeat knowledge, Cowork for delegated file-and-desktop work, and Chrome when the task must read, navigate, or act on websites. Use RunFrame skills whenever the work must follow a captured workflow.

Is Claude in Chrome safe to use on company systems?

It can be useful, but it is a beta browser agent and should start with trusted sites, narrow permissions, Ask before acting, and human review for financial, personal, customer-impacting, or work-critical actions.

Can Claude connectors send emails or change records automatically?

Connector behavior depends on the connector and permissions. RunFrame guidance is to require review before Claude sends external messages, changes records, deletes data, schedules meetings, or modifies permissions.

How do we make Claude outputs less generic?

Give Claude the real source material, examples, team voice, output format, decision rules, and review criteria. If the task repeats, turn those instructions into a Project, style, RunFrame skill, or Claude skill package.

Is Claude Design production-ready code?

Treat Claude Design as an exploration and handoff tool. It is useful for prototypes, decks, one-pagers, and software experience directions, but implementation still needs product, design, accessibility, and engineering review.

Where do I manage billing after setup?

Stripe checkout is live during onboarding. For invoice questions, seat changes, or subscription changes during early launch, email support@runframe.ai.

Troubleshooting

Issue

Likely cause

Next step

Google authorization fails

OAuth redirect mismatch, missing consent, or expired login session.

CSV import fails

Missing required columns, duplicate emails, or manager_email does not match an employee email.

Use columns name, email, title, role_category, manager_email and make one row top-level with blank manager_email.

Provisioning is blocked

Stripe webhook has not confirmed payment or the organization is not marked active/trialing.

Wait a minute after checkout return. If blocked, contact urgent support with the billing email.

Employee did not receive invitation

Email filtering, typo in employee email, or Resend delivery issue.

Confirm the employee email in the org chart, check spam, then resend the invitation or contact support.

Interview call was not placed

GHL booking email did not match the employee, no phone number was collected, or scheduler delay.

Confirm the employee exists, the booking email matches, and allow up to 5 minutes after the scheduled time.

Loom submitted but request still shows open

The Loom was submitted without the matching request ID or processing is still running.

Refresh the capture page. If still open, submit again from the request card or contact support with the request ID.

Contact Support

General help

support@runframe.ai

Target response: one business day.

Urgent blockers

urgent@runframe.ai

Use for provisioning, billing, or outage blockers.

Security questions

security@runframe.ai

Use for data handling and PII questions.