Unlock the consulting pipeline and view your Pharos AI teammates. Enter your admin key below.
Your team of specialists. Click any card to explore its infrastructure, read its full history, or queue a real run from the directive console.
Click any option to open it. Click the background to collapse back down.
Outside auditors that check the team's work and its process. They are not part of the team, and they report to you on their own channel, never through the Chief-of-Staff. Click a card for its full detail.
Each card opens the plain-language story of one engine that makes your team smart. This is the backstage view: what runs where, why it is fast, and how it learns you.
Every outside service and internal capability the team uses: what it does, whether its key is live, the free-tier ceiling, what it costs, and which teammates call it. Internal endpoints are tools your team owns; each wraps one or more outside providers so any agent can reach a capability without new wiring.
Every paid model call the team makes, rolled up. Local Qwen calls are free and are not counted here, so this is your true Anthropic spend. Amounts in USD.
This is the operating manual for your team. It explains, in plain language, what each teammate does, how a lead travels from "company we have never heard of" to "signed agreement," why the system exists, what it took off your plate, and exactly what it costs to run. No jargon required.
Running outreach by hand has a leak: the boring-but-critical steps (chasing follow-ups, fact-checking claims, tracking who went quiet) are exactly the ones a busy person drops first. That is what killed momentum before. The team exists to make those steps automatic and reliable, so the pipeline keeps moving even on the days you are heads-down on client work.
Before: you personally found companies, researched each one, wrote every email, checked every fact, remembered every follow-up, triaged every reply, prepped every call, and drafted every proposal. The pipeline only moved when you had time.
After: the team does the finding, drafting, checking, chasing, and prep. You do the judgment: approve a draft, take the call, make the final call. The pipeline moves whether or not you had a free hour.
A lead moves through four levels. Each teammate hands its output to the next, like a relay.
| Teammate | Level | What it does for you |
|---|---|---|
| Chief-of-Staff | Lead | Reads the whole team's work and gives you one daily briefing. Surfaces only what needs you. |
| Researcher | L1 | Scans the market daily for companies showing buying signals and writes a weekly read on where the opportunity is. |
| Prospector | L1 | Confirms a company is a real fit, verifies a public signal, and writes the research profile. |
| Personalizer | L2 | Writes the cold email and warm close in your voice, tuned to each company's gap. |
| Fact-Auditor | L2 | Checks every factual claim against its source and blocks anything unverifiable before it reaches a prospect. |
| Pipeline-Monitor | L3 | Watches every deal and your inbox daily and flags what needs action: follow-ups due, prospects gone quiet, new leads. |
| Sequencer | L3 | Drafts the follow-up emails that are due, in your voice, and queues them for your review. |
| Reply-Handler | L3 | Sorts inbound replies (interested, question, not a fit, spam) and drafts a warm first reply to the real ones. |
| Closer-Prep | L4 | Turns a prospect profile into a one-page discovery-call brief. |
| Proposer | L4 | Turns your call notes into a finished statement of work. |
AI models charge by the "token," which is roughly a chunk of a word the model reads or writes. Two kinds of model run this team:
The rule of thumb we follow: free local model first, pay for Claude only where it clearly earns its keep (your voice in client-facing copy, or judgment where a mistake is costly).
This is a review of every teammate using paid Claude today and whether the free local model could do the job instead.
| Teammate | Runs on today | Could Qwen do it? | Recommendation |
|---|---|---|---|
| Researcher | Qwen daily + Claude weekly | Daily yes; weekly is a judgment call | Keep as is. The free model already does the daily scan; the weekly "where is the opportunity" call is worth Claude. |
| Prospector | Claude | Partly | Hybrid Let Qwen draft the profile; keep Claude for the fit decision so we do not chase bad-fit companies. |
| Personalizer | Claude | No | Keep Claude. This is your voice going to a real prospect. Quality is the product. |
| Fact-Auditor | Claude | No | Keep Claude. This is the guardrail that catches wrong claims. Do not weaken it to save pennies. |
| Pipeline-Monitor | Code + Qwen | Already free | No change. Already optimal. |
| Sequencer | Tiered (Qwen / Haiku / Sonnet) | Already smart about it | No change. Uses the cheapest model that can hold your voice. |
| Reply-Handler | Qwen sort + Claude draft | Already split | No change. Sorting is free; only the reply draft pays for voice. |
| Closer-Prep | Claude | Yes | Move to Qwen. It is an internal brief only you read, so the cost of a small miss is low. Good free-model candidate. |
| Proposer | Claude | Mostly | Hybrid Let Qwen fill the statement of work; an optional light Claude polish before it goes out. |
| Chief-of-Staff | Qwen daily + Claude weekly | Likely yes | Try Qwen weekly too. The numbers are calculated in code; the model only phrases them, which the free model handles well. |
How the team actually brings in clients, stage by stage: who does what, what runs on its own, what you trigger, and exactly how to start. This is the single reference for the whole pipeline.
You only ever talk to the Chief-of-Staff. You never open another teammate's card to make it work. You tell the Chief-of-Staff what you want or what happened, in plain language, and it works the team and brings results back to you. Every output lands in your review queue. Nothing reaches a real person without your approval.
"Auto" runs on a schedule without you. "You trigger" means you ask the Chief-of-Staff, or it prompts you first.
| Stage | Teammate | Auto or you | What comes back to you |
|---|---|---|---|
| 1. Pick target companies | You (+ Researcher suggests) | You decide | You bring the targets. Researcher suggests sectors and a few companies; auto-discovery from market feeds is not built yet. |
| 2. Profile a company | Prospector | You trigger | A research profile of a company you name (a document to review, not a pipeline entry yet) |
| 3. Write the first cold email | Personalizer | You trigger | A personalized, fact-checked cold email draft |
| 4. Send the email | You | Manual | It sends as you, lands in your Sent |
| 5. Send follow-ups | Sequencer | Auto (when due) | A follow-up draft, ready to send |
| 6. Watch the pipeline | Pipeline-Monitor | Auto (daily) | Flags who is due, who went quiet |
| 7. A prospect replies | Reply-Handler | Auto (every 5 min) | The reply in your Replies tab + a suggested response |
| 8. Fact-check outbound | Fact-Auditor | Auto (built in) | Drafts arrive already fact-checked |
| 9. Prep for a discovery call | Closer-Prep | You trigger | A one-page brief: likely pain, questions to ask |
| 10. Write the proposal | Proposer | You trigger (it prompts you) | A filled statement of work, in your queue |
| Across all of it | Chief-of-Staff | Auto + always-on chat | One daily brief, plus nudges when something needs you |
This is the seam between research and the automated machine, and it is manual today. The team only works prospects that exist as records (with a contact email and a status). A company becomes one of those in two ways:
| When | What runs | What it does |
|---|---|---|
| Every morning, 8:00 AM | The whole team | Researcher, Pipeline-Monitor, Reply-Handler run; Sequencer drafts due follow-ups; Chief-of-Staff writes the daily brief |
| Every 5 minutes | The runner | Picks up what you queued in chat, pulls new email replies and pre-drafts responses, pushes stage nudges |
| Always on | Chat listener | Instant answers when you message the Chief-of-Staff |
| Mondays, 8:30 AM | Chief-of-Staff weekly | A higher-altitude weekly read |
Automated, no action from you: daily market research, watching who is due and drafting the follow-up, pulling in replies and pre-drafting a response, fact-checking every prospect-facing draft, the daily brief and live "what needs you" list, and advancing a prospect's stage when you tell it what happened.
You decide and you trigger: which companies to target (you ask Prospector), approving and sending every email (nothing auto-sends), asking for a cold email, call prep, or a SOW, and telling the Chief-of-Staff what happened so it can move a prospect forward.
The true end to end only completes when a real prospect replies. But the entire front half can be run today, with no prospect needed.
Do this in order. Each is one message to the Chief-of-Staff unless noted.