# graph8 for AI Agent Operator

> graph8 field guide for AI Agent Operator. Live URLs, real surfaces, screenshots from a public-data showcase tenant.

Source: https://docs.graph8.com/roles/agent-op/

*A field guide for AI Agent Operator. Live URLs, real surfaces. Screenshots captured from a showcase tenant with public-company data only.*

## The five things to use today

### 01. The agent roster — 8 agents across Agent + Twin entities

*5 Agents + 3 Twins. Agents use preset voices (David / Michael / Olivia). Twins use voice-cloned UUIDs of real humans (Ari, Din, Nick). Each has role, status, voice, phone, calendar appointment.*

**Open it at**

- `app.graph8.com/agents` — full roster

![Agents grid showing 8 agents: AI Receptionist, 10k free signup (David Male +1 (555) 010-0001), Hyperpersonalised (Michael Male), Graph8 Chat Agent (Olivia Female +1 (555) 010-0002), Graph8 Support Agent (CSM), Ari (Twin SDR Custom Voice), Din (Twin SDR), Nick (Twin AE Custom Voice) with columns Name/Entity/Role/Status/Phone/Voice/Appointment Event/Company Knowledge/Actions](../../../../assets/roles/agent-op/1-agents-list.webp)

***Fig 1.1** — `/agents`. Columns: *Name · Entity · Role · Status · Phone · Voice · Appointment · Company Knowledge.* All 8 Active.*

Agent · Receptionist AI Receptionistroutes inbound calls · phone directory · scheduling

Agent · SDR 10k free signup (David, Male)+1 (555) 010-0001 · trial-signup outbound

Agent · SDR Hyperpersonalised (Michael, Male)hyper-personalized outreach sequences

Agent · SDR/Chat Graph8 Chat Agent (Olivia, Female)+1 (555) 010-0002 · website visitor chat · demo booking

Agent · CSM Graph8 Support Agentpost-sale support, escalation

Twin · SDR AriCustom voice clone · UUID 8f5bc08c-… · dialer-driven

Twin · SDR DinCustom voice clone · many dial sessions

Twin · AE NickCustom voice clone · AE close-stage calls

### 02. Agent detail — Overview · Persona · Identity · Skills · Collections · Activity

*6 sub-tabs per agent. Configure mission, personality sliders, voice/phone, attached skills/workflows, knowledge collections, and per-agent activity log.*

**Open it at**

- Click any agent in `/agents` roster

![Graph8 Support Agent (CSM) Overview tab showing name, role CSM, Active badge, Company Knowledge button, Phone/Voice/Calendar (Appointment #0), 4 KPI tiles (Total Calls/Avg Duration/Success Rate/Collections=1), Description, Inbound/Outbound Call Instructions cards, Voicemail Message. Tab nav: Overview, Persona, Identity, Skills, Collections, Activity](../../../../assets/roles/agent-op/2-agent-detail.webp)

***Fig 2.1** — Graph8 Support Agent Overview. **6 tabs:** Overview · Persona · Identity · Skills · Collections · Activity. KPIs (Total Calls · Avg Duration · Success Rate · Collections) — empty until agent runs.*

> **Tip: Try this now**
> **Operator move.** Per agent monthly: re-read the Persona prompt aloud. Adjust the 3 sliders if customer feedback says the agent is too pushy / too formal / too verbose. Sliders are blunt but effective.

### 03. Skills + Workflows + Company Knowledge

*Attach reusable skills (action recipes) and workflows to agents. Plus Company Knowledge (RAG corpus from the 46 brand DNA docs). Today: 0 skills attached across agents — easy to populate.*

**Open it at**

- Agent detail → Skills sub-tab
- Agent detail → Collections sub-tab
- Top of agent detail → Company Knowledge button

![Skills tab for Graph8 Chat Agent showing Agent Skills & Workflows header, Attach Skills button, empty state 'No Skills or Workflows Attached', Skill Usage Statistics row at bottom (Total Skills=0, Executed 30d=0, Success Rate=0%)](../../../../assets/roles/agent-op/5-agent-analytics.webp)

***Fig 3.1** — Skills tab. **0 skills attached** across all agents today. Surface ships, ready for population. *Skill Usage Statistics (Total · Executed 30d · Success Rate)* all at 0.*

> **Tip: Try this now**
> **Operator move.** Populate Skills for the agents that get the most traffic (Graph8 Chat + 10k free signup). Add reusable skills like "book a meeting" · "qualify the lead" · "send a calendar invite" · "create a CRM contact." Track Skill Usage to see which fire most + which fail.

### 04. Twin agents — voice-cloned versions of real reps

*Twins differ from Agents: custom voice UUIDs (cloned from a real human), no Delete button (less freely removable), voicemail scripts written in first-person human voice. Today: 3 Twins (Ari/Din/Nick) covering SDR + AE roles.*

![Ari Twin agent detail showing Active, SDR role, Voice with UUID `[voice-clone-id]` (custom voice clone), no standard phone, Voicemail Message populated with personalized script 'this is Ari from Graph8'](../../../../assets/roles/agent-op/7-twin-agent.webp)

***Fig 4.1** — Ari Twin agent. Voice = UUID `8f5bc08c-2c5a-…` (cloned human voice). Voicemail script in first-person: *"\{First_Name\} this is Ari from Graph8…"* **brand rule violation** — "Graph8" should be lowercase. Fix in voicemail textbox.*

> **Tip: Try this now**
> **Operator move.** Audit each Twin voicemail script monthly for brand rule violations ("graph8" lowercase) + tone drift. Re-record the voicemail if the script has changed. Then re-run a small test dial session to verify the new clone hits the right notes.

### 05. Monitoring — Inbox + Dialer + Copilot

*Replies hit the shared Inbox tagged by sequence (not by agent — see caveat). Dialer sessions show Twin-driven campaigns. Copilot can pull weekly agent performance — when the dialer analytics endpoint works (currently broken with HTTP 422).*

![Inbox Focused tab with 6 threads (a contact 'kill the tool sprawl', a contact 'how AI companies are using this' with Not Interested + Draft tags, a contact, a contact, a contact, a contact) tagged by SEQUENCE source (REVOPS_SAAS, SALESLEADERS_AI&ML)](../../../../assets/roles/agent-op/9-dialer-voice-ai.webp)

***Fig 5.1** — Inbox. Replies tagged by SEQUENCE source. **Heads-up:** no explicit "Handled by Olivia" agent-attribution tag — replies land in shared inbox.*

![Copilot panel response to 'How are my AI agents performing this week?' showing pulled agent list with table (10k free signup → Conversion → positive-interest responses + signup; Graph8 Chat Agent → SDR/Chat → Website visitor engagement demo booking), Performance Data section noting dialer analytics endpoint returned HTTP 422 error, offers alternatives](../../../../assets/roles/agent-op/10-copilot-agentop.webp)

***Fig 5.2** — Live copilot. Q: *"How are my AI agents performing this week?"* Returns agent role table + flags **dialer analytics endpoint HTTP 422**. **Real bug** — Operator should escalate to engineering.*

> **Tip: Try this now**
> **Operator move.** Weekly cycle: check `/dialer` Sessions for Twin/Agent call volume + status (Paused vs Active). Check Inbox for sentiment trends per sequence. Ask copilot *"Which agent had the most positive replies this week?"* — when the analytics endpoint is healthy, you get a real answer.

## URL cheat-sheet

| To do this | Go here |
| --- | --- |
| Agent roster (8 agents) | `/agents` |
| Agent detail (6 sub-tabs) | Click any agent row |
| Persona prompt + sliders | Agent → Persona tab |
| Voice + phone + calendar | Agent → Identity tab |
| Skills + Workflows attached | Agent → Skills tab |
| Knowledge collections | Agent → Collections tab |
| Per-agent activity log | Agent → Activity tab |
| Dialer session monitoring | `/dialer` |
| Inbox (replies tagged by sequence) | `/inbox/email` |
| Brand DNA docs (RAG source) | `/studio?mode=global` |
| Open Copilot | Cmd+Shift+K |

## Heads-up — Operator gaps + bugs to escalate

> **Caution:**
> **1. **BUG** Dialer analytics endpoint returns HTTP 422 to Copilot.** Copilot can't pull weekly agent performance. Escalate to engineering — this blocks the natural-language "how are my agents performing?" question that Operators ask most.
>
>  **2. **BRAND** Ari Twin voicemail script has "Graph8" (capital G).** Should be lowercase "graph8" per brand rule. Audit all 3 Twin voicemails + agent prompts for this.
>
>  **3. Skills attached: 0 across all agents.** Surface shipped but unused. Populate skills + workflows on the top-traffic agents first.
>
>  **4. Inbox doesn't attribute replies to a specific AI agent.** Threads tagged by sequence only — no "Handled by Olivia" tag. Workaround: trace by phone number + timestamp for now.
>
>  **5. Activity tab is "Nothing here yet" across all agents.** Per-agent activity log surface exists but isn't populated. Either no activity is being written to it, or activity is being written elsewhere — confirm with engineering.
>
>  **6. Only 2 agents have phone numbers assigned.** 10k free signup (+1 (555) 010-0001) and Graph8 Chat Agent (+1 (555) 010-0002). Other 6 agents are voice/chat-only. Assign phones to whichever agents should take inbound calls.

---

<small>Generated 2026-05-20 against `app.graph8.com` (org `your org`). Surface state verified live</small>