Last updated: 2026-03-21

Glance Docs

Glance docs help new tenants get set up quickly, connect Claude, and run the Brand Interview without having to guess how the workflow works.

Who This Is For

• New tenants who need to log in and start using the product
• Teams moving from ChatGPT to Claude
• People running the Brand Interview through the MCP connector

Where To Read This Guide

You can read these docs in two places:

• in the git repo under docs/user-guide/
• on the Glance docs site at docs.getglance.com

Start Here

1. Read Start Here
2. Read Connect Claude
3. Read Brand Interview
4. Read Migrating from ChatGPT
5. Review Troubleshooting if anything is unclear

What To Expect

The Brand Interview should:

• ask one question at a time
• offer A/B/C options when there is enough information to recommend them
• save the full conversation transcript
• generate BRAND.md, DESIGN.md, and a tenant preview URL

If you want Claude to think with you, not just answer prompts, use the recommendation and option flow. That produces better outputs than one-word answers.

Last updated: 2026-07-01

Start Here

This is the fastest path to using Glance with Claude.

1. Sign in with your magic email link.
2. Add the Glance connector to Claude.
3. Authenticate the connector.
4. Start the Brand Interview.

If you are new to Claude, read this page first. If you are moving from ChatGPT, the next section explains the differences that matter most.

What You Need

• Your tenant login email
• Claude desktop or Claude web
• The Glance connector URL for your tenant

Magic Email Login

Use the same email address your team gave you. Open the magic link from your inbox and return to the tenant workspace after the browser confirms login.

If the link expires, request a new one. Do not keep trying an old link.

Add The Connector

In Claude, open Settings, then Connectors or Integrations, then add a custom connector. Paste the Glance connector URL and save it.

Claude will then open the authorization flow. Approve it and return to Claude after the browser shows success.

Confirm It Worked

You should see the connector as connected before starting the interview. If the wrong tenant appears, stop and reconnect before proceeding.

Last updated: 2026-07-01

Brand Interview in Claude

This guide shows tenants how to use the Brand Interview connector inside Claude. The goal is to collect brand context in a real conversation, save the transcript, and generate the tenant's brand documents and preview for review.

Back to main guide

What You Need First

• Access to your tenant workspace
• A verified email address for magic email sign-in
• Claude desktop or Claude web access
• The Brand Interview connector URL for your tenant

If your team gave you a connector URL, use that exact URL. In most cases it will be https://connector.getglance.com/mcp.

1. Sign In With Magic Email

1. Open your tenant login page.
2. Choose email sign-in.
3. Enter the email address your team has on file.
4. Check your inbox for the magic link.
5. Open the link to finish signing in.
6. Return to the workspace after the browser confirms you are logged in.

Tips:

• Use the same email address you will use to connect Claude.
• If the link expires, request a fresh one rather than reusing an old email.
• If you manage more than one tenant, be careful to sign into the right one before continuing.

2. Add The Connector To Claude

1. Open Claude.
2. Go to Settings, then Connectors or Integrations.
3. Choose Add custom connector.
4. Paste the connector URL your team provided.
5. Save or continue.
6. Wait for Claude to open the authorization flow.

If Claude asks for an account, use the same email address you used for tenant sign-in.

3. Authenticate The Connector

1. Approve the connector when Claude opens the browser authorization page.
2. Confirm that the workspace or tenant shown in the browser is the right one.
3. Complete any consent screens.
4. Return to Claude and wait until it shows the connector as connected.

If you see a different tenant than expected, stop and reconnect before you start the interview.

4. Start The Conversation

Once the connector is connected, you can begin with:

start brand interview

You can also say:

I want to do a brand interview for <company name>.

Claude will then begin asking questions one at a time. The interview may use multiple turns, examples, follow-up questions, and option lists when that helps it collect better context.

5. How To Get The Best Results

The interview works best when you treat Claude like a thinking partner, not a form you are filling out.

Be concrete

• Give real examples instead of broad categories when you can.
• If Claude asks for a website, share the actual URL.
• If it asks for a feeling, story, audience, or proof point, answer with one specific instance instead of a slogan.

Use the options when they appear

Sometimes Claude will offer A/B/C choices and a recommendation.

• Pick the closest option if one fits.
• If none fit, answer in your own words.
• If you choose an option letter, Claude records the full meaning of that option, not just the letter.

Let the conversation breathe

• It is okay to ask Claude questions back.
• It is okay to explore a side idea before returning to the interview.
• If the conversation goes off track, just say back to the interview when you are ready to continue.

Ask Claude to think with you

Good prompts include:

• "Help me compare a few directions."
• "What would you recommend and why?"
• "Can you give me options and tell me which one you prefer?"
• "What am I missing?"
• "Can you summarize what we know so far?"

Review before finalizing

At the end, Claude will present the draft outputs for confirmation. Look for:

• the brand summary
• the design direction
• any questions or assumptions that still need edits
• the preview URL
• the generated Markdown files

If something is wrong, ask for edits before you approve the final output.

6. What You Should Expect At The End

When the interview is complete, Claude should show:

• the saved brand interview transcript
• the generated BRAND.md
• the generated DESIGN.md
• a tenant-isolated preview URL you can open in the browser

Those artifacts are meant to be reviewed by humans. The preview URL should be web-accessible for your tenant, not a local machine link.

If the preview or documents do not look right, ask Claude to revise the relevant section and regenerate the output.

Troubleshooting

The connector says it is not connected

• Re-open the connector settings in Claude.
• Confirm you authorized the right connector URL.
• Sign out and reconnect if you used the wrong tenant email.

The interview feels too short

• Ask Claude to go deeper on the current topic.
• Share a concrete example, URL, or story.
• Ask for options if you are unsure how to answer.

Claude drifts away from the interview

• Say back to the interview.
• Then answer the question directly or ask Claude to restate it.

You want to revisit the final output later

• Ask for the saved transcript.
• Ask for the generated Markdown files.
• Ask for the preview URL again.

Suggested Prompt

If you want a simple starting message, send this in Claude:

Start the brand interview for <company name>. I want help thinking through the brand, and I’m happy to answer one question at a time.

If you want Claude to act more like a collaborator, send:

Start the brand interview. Ask me the questions one at a time, offer options when you have enough information, and help me think through the tradeoffs.

Last updated: 2026-03-21

Marketing

Your marketing agents handle everything from campaign orchestration to email copy to webinar presentations. Campaigns run through a deterministic pipeline called Lobster — not freeform agent improvisation. You remain the final decision-maker on anything that reaches the outside world.

Back to main guide

Your Marketing Agents

Conductor (Campaign Orchestrator) — Launches campaign workflows through the Lobster pipeline engine. Conductor does not write copy, generate HTML, or coordinate specialists manually. It gathers your campaign inputs (topic, dates, slug), launches the workflow file, and narrates the approval gate. Lobster owns the execution; Conductor owns the conversation with you.

Quill (Copywriter, Head of Content) — Writes all marketing copy across the agency: landing pages, email sequences, SMS campaigns, webinar scripts, ads, and social posts. Quill follows the active brand voice, selects a copywriting persona per project (stated explicitly before drafting), and runs every piece through the humanizer before delivery. Reports to the CMO.

Forge (Email Engineer) — Builds production-grade HTML emails using React Email and Resend. Forge v3 uses deterministic rendering from design system tokens — not agent-generated TSX. Every template is tested across Apple Mail, Gmail, Outlook, Yahoo, and dark mode variants before it reaches you. Reports to the Frontend Engineer.

Atlas (Strategist) — Designs conversion funnels, runs CRO audits, manages analytics tracking, and plans campaign strategy. Atlas coordinates the strategic layer — what to build and why — while Conductor handles the execution pipeline.

Prism (Webinar Architect) — Creates webinar presentations and manages the full webinar pipeline from ideation through live event and replay.

Brand Designer (Forma) — Creates visual design specifications: color palettes, typography, layout guides. Works to WCAG 2.2 AAA accessibility standards. Design specs only — never writes code.

How Campaigns Work (Lobster Pipeline)

Campaigns run through a deterministic pipeline engine called Lobster. This replaces manual agent coordination with a repeatable, governed workflow.

The Flow

1. You brief Conductor — Provide the topic, event date, launch date, campaign slug, and page slug. Conductor gathers any missing inputs.
2. Conductor launches Lobster — The workflows/campaign-webinar.lobster workflow file runs. Conductor never inlines pipeline steps or composes its own orchestration.
3. Lobster coordinates specialists — The pipeline spawns Quill (copy), Forge (email HTML), and other specialists in the right order, with handoff contracts between each stage.
4. Brand gate — Before approval, deliverables pass through the brand quality check (CCO vertical). The Creative Director reviews design, and the QA Engineer runs adversarial tests.
5. You approve — Conductor presents the completed assets and waits for your explicit go-ahead.
6. Publish — After approval, the pipeline publishes and creates campaign records.

If the workflow fails at any stage, Conductor reports the failure and stops. No manual fallback.

Webinar Campaign Templates (16 total)

A full webinar campaign generates 16 email templates through the Lobster pipeline:

Paperclip Integration

You can also trigger campaigns from the Paperclip dashboard. Assign a campaign task to Conductor, and the lifecycle bridge routes it through the same Lobster pipeline. Progress and governance records are visible in Paperclip throughout.

How SMS Campaigns Work

SMS campaigns run through Twilio alongside email campaigns. The same approval rules apply — no SMS goes out without your sign-off. SMS content follows the same brand voice rules and humanizer pass as email copy.

How the Webinar Pipeline Works

Prism manages webinars through a structured sequence:

1. Ideation — You provide the topic and audience. Prism develops the concept using the Perfect Webinar framework.
2. Presentation — Prism generates an HTML presentation. You review and approve.
3. Registration — The Lobster pipeline builds the registration funnel (landing page, email confirmation, reminders).
4. Promotion — Quill writes the promotional emails. The brand gate reviews for voice. You approve before send.
5. Live event — Zoom integration handles recording.
6. Replay — After the event, the zoom-recording skill activates the replay page so registrants can watch on-demand.

Brand Guidelines Enforcement

Every marketing agent follows the active brand system automatically:

• Voice: Loaded from your brand guide (built during the brand interview)
• Kill list: Words like "sacred," "miracle," "therapy," "spirit guide," and "breakthrough" are automatically flagged and rewritten
• Health claims: "Heal," "cure," "treat," and "guaranteed results" are prohibited in all marketing content
• Humanizer: Before any copy reaches you or an external audience, it passes through a humanizer that strips AI writing patterns
• Brand gate: The CCO vertical (Creative Director + QA Engineer) reviews deliverables for brand consistency before they reach you for final approval

You do not need to enforce these rules manually — they are built into every agent's operating instructions.

Quality Gate (QA Engineer)

Before any campaign email reaches you for final approval, it passes through the QA Engineer (reporting to the CCO). This runs automatically as part of the pipeline.

What the QA gate checks:

• Responsiveness — Renders at mobile, tablet, and desktop viewports, plus dark mode and images-off variants
• Gmail clipping — Checks whether the email exceeds Gmail's 102KB clip threshold
• Deliverability score — Estimates inbox placement based on text-to-image ratio, link count, and marketing-trigger phrases
• Link health — Verifies every link resolves correctly
• Regression — Compares scores against previous sends
• Accessibility — WCAG AA check (alt text, color contrast, font sizes)
• Brand gate — Voice, kill list, CTA alignment, humanizer pass

What Needs Your Approval

• All outbound emails (marketing, nurture, campaign)
• Webinar presentation content
• Landing page copy and design
• SMS messages
• Any content published to external channels

What Happens Automatically

• Lobster pipeline execution (specialist coordination, handoff contracts)
• Brand voice checking across all drafts
• Humanizer pass on all copy
• Email template rendering and cross-client testing
• Bounce handling and list hygiene
• Campaign performance logging

Back to main guide

Last updated: 2026-03-21

Sales

Your sales system is built around the Sales Director agent — the orchestrator that manages the full pipeline from first inquiry to retreat booking. The director coordinates a team of specialized sub-agents and keeps everything tracked in Attio CRM. Lumina, the conversational sales agent, handles direct prospect conversations via iMessage and SMS. The guiding principle: create clarity and hold space for the right decision. No pressure, no manufactured scarcity, no shame-based messaging.

Back to main guide

Your Sales Agents

Sales Director — The primary orchestrator. Responds to new inquiries within 5 minutes, personalizes follow-ups using call transcript data from Fireflies, manages deals in Attio, and coordinates the sub-agents below. Sends from your brand's configured sending address.

Lumina (Conversational Sales) — Handles direct prospect conversations via iMessage and SMS. Lumina classifies intent, qualifies interest, answers FAQs, and when a conversation reaches the right point, warmly invites the lead to continue on iMessage. Hands off to you or the sales team when things get serious — crisis mentions, payment questions, or high-intent purchase signals always route to a human.

Enrollment Agent — Handles the enrollment workflow once a lead is ready to commit. Works with Attio, Typeform, and Resend to move people through the enrollment steps.

Lead Nurture Agent — Manages longer-term follow-up for leads who are not ready to commit yet. Re-engages on meaningful events (new retreat dates, a relevant article) — never on arbitrary timers.

Coach Agent — Supports the sales process with coaching insights. Coordinates through Slack and Attio.

Pricing & Yield Agent — Manages pricing strategy, payment plans, and availability. Works with Attio and WeTravel to keep pricing and seat counts accurate.

Lumina's Intelligence Sub-Skills

Lumina uses 11 specialized sub-skills to handle conversations intelligently:

Compliance Guards

Every outbound message from Lumina passes through seven compliance guards before dispatch:

If any guard fails, the message is blocked and you are notified. Guards are fail-closed — if a guard cannot determine compliance, the message does not send.

How the Pipeline Works

The sales director responds to four key triggers:

1. Lead Capture — When someone signs up via newsletter or form submission, the director creates an Attio contact (stage: New Lead), sends a welcome email with a Calendly link within 5 minutes, and logs the touchpoint.

2. Post-Call Follow-Up — After a discovery call, the director reads the Fireflies transcript, personalizes a follow-up email with specific details from the conversation, and updates the Attio deal record.

3. Nurture Sequence — For leads who need more time, the director moves them into a nurture track. Follow-ups are triggered by meaningful events, not calendar timers.

4. Closing — When a lead is ready, the director drafts the proposal. But closing actions (deposit requests, contract links) always require sign-off from the assigned rep — Ethan or Val — before proceeding.

CRM Integration (Attio)

Every interaction is tracked in Attio:

• New leads get a People record created automatically
• Every email send is logged as a note on the contact record (via the resend/log-to-attio skill)
• Deal stages are updated as leads progress through the pipeline
• Fireflies call transcripts are linked to the contact record
• Attio webhook sync keeps records current when external changes happen (Squarespace orders, form submissions)

When the sales director processes any pipeline trigger, the first step is always reading the Attio contact record and all available Fireflies notes. No outreach happens without full context.

Email and iMessage Follow-Up

Email is used for proposals, documentation, and nurture sequences. Every outbound email is CC'd to the assigned sales rep.

iMessage is used only for warm, informal touchpoints — and only when the lead has explicitly opted into text communications (TCPA compliance). Lumina will never text someone who has not opted in.

SMS follows the same compliance rules as iMessage. All SMS sends route through the compliance guards listed above.

Approval Workflows

No outbound communication of any kind goes out without explicit approval. The agent drafts; a human decides.

Safety and Compliance

The sales agents follow strict rules to protect your brand's integrity:

• No health claims — "Heal," "cure," "treat," and "guaranteed results" are prohibited
• No pressure tactics — No countdown timers, no manufactured scarcity, no shame-based messaging
• No medical data in outreach — Medical intake information is never included in outbound messages (HIPAA)
• Plant medicine care — All references to plant medicine offerings are handled with heightened legal awareness; no medical or therapeutic claims, ever
• Objection handling — Uses Feel-Felt-Found framework. Always probes once before accepting "I need to think about it" at face value

What Happens Automatically

• First-response within 5 minutes of new inquiry
• Attio contact creation and deal tracking
• Send logging to CRM after every outbound email
• Fireflies transcript reading before follow-ups
• Retreat date and seat count fetching from Airtable
• Compliance guard validation on every outbound message
• Lumina conversation intelligence (intent classification, lead qualification)

Back to main guide

Last updated: 2026-03-21

Programs

The programs agents manage the participant journey from first application through retreat preparation. This is the most sensitive domain — it handles medical information, screening decisions, and clinical safety. Every step is designed with care, and you have final say on all critical decisions.

Back to main guide

Your Programs Agents

Onboarding — The front door. Receives new applications, coordinates the initial intake process, and routes participants to the right next step. Works with Attio (CRM), Typeform (forms), Zoho Sign (agreements), Resend (email), and Slack (team coordination).

Integration — Manages the post-screening integration process. Coordinates between clinical review outcomes and program enrollment, ensuring approved participants move smoothly into preparation.

Preparation — Once someone is approved, this agent manages the preparation process: logistics information, pre-retreat materials, and readiness check-ins. Works with Retreat Guru for scheduling and Attio for tracking.

Retreat — Handles retreat-specific coordination: scheduling, seat management, and retreat-day logistics. Works with Retreat Guru and Airtable for dates and availability.

The Onboarding Flow

Here is how a new participant moves through the system:

1. Application Received

When someone submits an application through Typeform, the onboarding agent picks it up. It creates or updates their Attio record and confirms receipt via email.

2. Application Screening

The application is reviewed for completeness and flags. You see a clear PASS / FLAG / HOLD recommendation.

3. Medical Screening

If the application moves forward, the medical intake form is reviewed. This is where the system is most cautious:

• SSRI and medication review — Checks for medications that may interact with the retreat experience. The approach follows your program's configured clinical framing (not a blanket requirement).
• Contraindication flags — Known contraindications are cross-referenced against the psychedelic safety knowledge base (Gbrain).
• Safety-first decisions — If anything is flagged, you receive the details and a recommendation. You always make the final call.

Medical information is treated with the highest sensitivity. It is never included in outbound emails, never shared with sales agents, and never surfaced in handoff payloads to other domains.

4. Approval and Agreement

Once you approve a participant, the onboarding agent coordinates the Zoho Sign agreement. The participant receives their agreement electronically, signs it, and the signed copy is tracked.

5. Preparation

After the agreement is signed, the preparation agent takes over:

• Sends pre-retreat information and logistics details
• Coordinates with Retreat Guru for scheduling and seat management
• Sends readiness check-ins as the retreat date approaches
• Updates the Attio record at each milestone

Integrations

What Needs Your Approval

• Every application screening decision (PASS / FLAG / HOLD)
• Every medical screening decision — no exceptions
• Participant acceptance or rejection
• Any communication about medical topics
• Agreement templates before first use

What Happens Automatically

• Application receipt confirmation emails
• Attio record creation and stage updates
• Medical form cross-referencing against the safety knowledge base
• Preparation timeline tracking
• Retreat date and availability lookups from Retreat Guru

A Note on Safety

The medical screening process is built with extra layers of caution. Email content that mentions medications, SSRIs, or contraindications is always routed to the onboarding agent — never to sales or marketing. All medical data is treated as protected information following HIPAA-aligned handling practices.

If the screening process encounters something it cannot confidently assess, it stops and brings it to you with full context. Borderline clinical decisions are never made autonomously.

Back to main guide

Last updated: 2026-03-21

Operations

The operations layer keeps the system running day to day. It handles inbox triage, knowledge sync, transcript ingestion, memory routing, outbound validation, and infrastructure. Most of this happens quietly in the background — you only hear from these agents when something needs your attention.

Back to main guide

Admin Docs And Internal References

The internal docs you read in the admin panel are sourced from this repo. In practice that means:

• user-facing operational guidance lives under docs/user-guide/
• developer/operator references live under docs/technical/
• the docs pages render those files at docs.getglance.com

If something in the docs is wrong, the source fix is a normal repo edit first, followed by a deploy of the web/ app.

Email Triage (Coordinator Agent)

The coordinator is your inbox monitor. Every 30-60 minutes, it checks your configured inboxes and sorts incoming emails by domain:

• Sales emails go to the sales director
• Onboarding emails go to intake coordination
• Marketing emails go to the marketing team
• Finance emails go to the payroll agent
• Everything else gets a general classification and routing recommendation

How Triage Works

1. The coordinator fetches new emails from both inboxes
2. Auto-replies, out-of-office messages, and bounce notifications are filtered out automatically
3. Each real email gets an Attio lookup — if the sender is a known contact, their stage (New Lead, Confirmed Participant, Alumni, etc.) influences routing
4. The email is classified by domain and handed off to the right agent
5. If the inbox is quiet (no actionable emails), the coordinator silently logs "all clear" without bothering you

The coordinator processes up to 20 emails per cycle. If there are more, the rest are queued for the next cycle.

Memory System

Lumina OS uses curated file memory, tenant Gbrain corpus retrieval, and hindsight for episodic relationship context.

Memory Routing

The source of truth for memory routing is config/memory-routing.yaml. This file determines how each agent retrieves context:

All agents use curated files as the universal memory path. Agents query Gbrain only when document-grounded retrieval is needed and use hindsight for episodic relationship context.

Conflict Resolution

When retrieved context and curated file memory disagree, agents verify the source before updating file memory. Do not let unverified retrieval override curated operating memory.

Failure Behavior

• Hindsight timeout: 5-second timeout, logged to daily memory, agent continues with file-memory context only
• Gateway search failure: Logged to daily session log, falls back to explicit file reads

Outbound Gate (ClawWrap)

Every agent-initiated external send — email, SMS, Slack, WhatsApp — routes through the outbound gate before dispatch. The gate validates targets, not permissions (your per-session approval is still required for every send).

How It Works

Key Rules

• Fail-closed: Unknown targets, unlisted audiences, and disabled channels are denied automatically
• No hardcoded addresses: Agent SOUL.md files and skills never contain email addresses, phone numbers, or channel IDs — the gate resolves them from targets.yaml
• New targets: Adding a new recipient (a new Slack channel, a new retreat WhatsApp group) requires an entry in targets.yaml and outbound-policy.yaml

Paperclip Lifecycle Bridge

Paperclip is the project management dashboard for task assignment and governance. The lifecycle bridge connects Paperclip issues to agent execution:

1. You assign a task in Paperclip to a specific agent (or the system routes it based on domain)
2. The bridge translates the Paperclip issue into an agent-compatible assignment brief
3. The agent executes and reports progress back through the bridge
4. Governance records (who did what, when, with what outcome) are visible in Paperclip

This is especially useful for agents with restricted autonomy — instead of the agent deciding what to work on, you assign tasks through Paperclip and the infrastructure handles the lifecycle.

Heartbeat System

Every agent runs on a heartbeat — a regular check-in cycle (typically every 30-60 minutes) where the agent reviews its task list and takes any needed action:

• If nothing needs doing, the agent logs a quiet "OK" and moves on. You will not hear from it.
• If there is work, the agent processes it and reports results through the appropriate channel.
• If something is wrong, the agent alerts you.

You do not need to manage heartbeats. They run automatically.

Knowledge Sync (Meridian)

Meridian keeps your knowledge base current by syncing documents from Google Drive into Gbrain. This runs daily and covers 13 knowledge areas (psychedelic safety, programs, marketing, sales, executive, finance, and more).

Meridian reads from Google Drive (read-only) and updates the local Gbrain knowledge areas. Other agents use this knowledge base when they need context — for example, the medical screening flow queries the psychedelic safety knowledge area when reviewing intake forms.

Observability (Opik)

All agent interactions are traced through Opik (comet-ml/opik-openclaw). This gives you visibility into:

• Traces and spans — What each agent did during a session, step by step
• Token costs — How much each agent session consumed
• Latency — Response times across the system

Opik runs on all environments (dev, staging, production). You do not need to configure it — it is enabled by default.

Infrastructure (Railway + Mac Mini)

The Mac Mini also serves as the iMessage relay for all environments (imsg-plus + SSH + Tailscale). Secrets are managed through Doppler with separate dev/staging/production configurations.

Each Glance tenant runs as its own isolated instance on this infrastructure.

Transcript Ingestion (Curator)

Curator processes your teaching transcripts and adds them to the knowledge base:

1. Drop a transcript into corpus/staging/ as a Markdown file
2. Curator picks it up on its next heartbeat cycle (every 30 minutes)
3. Four-stage pipeline runs automatically: triage (scored out of 20), format, organize, RAG sync
4. You get an iMessage summary after each run

Curator never deletes anything from the knowledge base without your explicit instruction. Incoming files are scanned for prompt injection attempts — flagged files are quarantined.

Organizational Config

One file drives all the constants your agents use: config/org.yaml. It holds your organization's name, domain, email addresses, Slack channel names, brand colors, Typeform IDs, and cleanup settings.

• When anything needs to change (new email, new Slack channel), edit one file and all agents pick it up automatically
• API keys and credentials stay in .env — never in the config file

What Needs Your Approval

• Invoice payment decisions (PASS / FLAG / HOLD)
• Transcript triage overrides
• Any knowledge base deletions
• Changes to outbound targets or policies
• Paperclip task assignments

What Happens Automatically

• Email triage and routing (every 30-60 minutes)
• Heartbeat monitoring across all agents
• Gbrain retrieval and hindsight recall when relevant
• Outbound gate validation (fail-closed)
• Google Drive to Gbrain knowledge sync (daily)
• Transcript processing pipeline
• Opik trace collection
• Auto-reply and spam filtering

Back to main guide

Last updated: 2026-07-01

Troubleshooting

The connector says it is not connected

• Re-open Claude settings and check the connector URL.
• Confirm you authenticated the right tenant.
• Sign out and reconnect if you used the wrong email.

The interview feels too short

• Answer with a specific example instead of a category.
• Ask Claude to go deeper on one part of the answer.
• Ask for options if you are not sure which direction to choose.

Claude drifted off-topic

• Say back to the interview.
• Ask Claude to restate the current question.
• Answer directly, then continue.

The preview or docs look wrong

• Ask Claude to regenerate the output.
• Ask it to show the saved transcript.
• Confirm the generated BRAND.md, DESIGN.md, and preview URL.

Appendix A · Connect Claude

Last updated: 2026-07-01

Connect Claude

This is the exact connector step for Glance.

Add The Connector

1. Open Claude.
2. Go to Settings.
3. Open Connectors or Integrations.
4. Add a custom connector.
5. Paste the Glance connector URL for your tenant.

Authenticate

Claude will open the authorization flow in your browser.

1. Sign in with the same email you used for the tenant.
2. Approve the connector.
3. Wait until Claude shows the connector as connected.

If the browser shows the wrong tenant, stop and reconnect before starting the interview.

What Good Looks Like

• The connector appears connected in Claude.
• The tenant shown in the browser matches your workspace.
• You can now start the Brand Interview.

Appendix B · Migrating from ChatGPT

Last updated: 2026-07-01

Migrating From ChatGPT

If you are used to ChatGPT, Claude will feel similar in some ways and different in important ways.

What Changes

• Claude is better used as a collaborator than as a one-shot answer machine.
• The Brand Interview may take multiple turns.
• Claude can offer options, recommendations, and follow-up questions.
• You can correct or redirect the conversation without starting over.

How To Use Claude Well

• Ask for options when you are unsure.
• Ask Claude what it recommends and why.
• Give concrete examples, URLs, names, and stories instead of abstract labels.
• Let Claude ask follow-up questions before forcing a final answer.
• Say back to the interview if the conversation wanders.

Memory And Context

If you have useful ChatGPT memory, notes, or previous drafts, bring the useful pieces over manually and use them as source material in Claude.

If you want to import memory directly, Claude's import flow is: https://claude.com/import-memory

Practical Advice

Start with one clear sentence about the company, then answer the interview one question at a time. If Claude offers A/B/C choices, choose the closest option or explain why none fit.