Agents Database
The live Ops DB every agent reads & writes — brands, influencers, deals, the queue tabs, and the Agents-at-Work log.
Operations manual · How every agent plays its part
The Battalion, Explained
Ten agents that run like an orchestra on your Hostinger VPS — orchestrated by Hermes, thinking with Opencode. Each waits for its cue, does one job, writes the result to the shared sheet, and hands the baton to the next. This page is the score — what triggers each one, what it produces, exactly how it works, and what to check when it doesn't.
FUP keeps every open thread warm · Hermes runs the battalion on your VPS · the Orchestrator is the conductor
How it all plays
Three things start the music; everything else is the agents passing the baton.
① You tick a stage
On the Influencer Working Sheet. An Apps Script turns the tick into a row in the Triggers tab — and the Gmail label on that creator's thread tells the agent which email to read.
② You click a button
On the website (brand side). Each click — Intake, Generate round, Contract signed, Generate invoice — drops a row in the Actions tab.
③ The clock ticks
Hermes on your Hostinger VPS. It drains both tabs every 10 min, runs follow-ups daily, and posts a health check each morning — no extra dashboard to babysit.
The baton, left to right. A creator replies with a rate → I-1 works out the CPM → you tick Pitch → I-2 adds them to the roster → a brand picks them → I-3 draws up the contract. On the brand side, intake → B-1 builds the folder + master sheet → Generate round → B-1.5 sends the shortlist → both sign → B-2 confirms the creator. Both lanes meet at DLV/INV, which chases the draft, closes on the live video, and bills the brand. FUP chases anyone who goes quiet. Nobody sends money, a contract, or a message without you approving it first — those are the 5 gates.
The database & the records
One Google Sheet — the Ops DB (16QKwPPb4MdO…JjPEk) — is the whole database, the audit trail, and the backup. Every tab below is one sub-sheet — and they’re all in your updated IMA_Database.xlsx, including the two queue tabs.
▸ The "Agents at Work" tab on the website = the Agent_Log tab
Every agent writes one line per action to Agent_Log. That feed is what the website's Agents-at-Work view reads — same six columns, one-to-one. This is your record + backup of everything the battalion does. Where to add it: create a tab named exactly Agent_Log inside the Ops DB and point the website's Agents-at-Work read at it.
Agent_Log headers → timestamp · agent · action · record_id · result · waiting_on_approval
| Tab | What it is | Header row (the columns) |
|---|
The two queue tabs — Triggers and Actions — are now in your updated IMA_Database.xlsx alongside the rest. Share the whole Ops DB with the service-account email as Editor so Hermes can read and write it. Opencode does the thinking; Google still stores the data — keep the Sheets, Drive, Docs & Gmail APIs enabled for that service account.
Troubleshooting
Start with the agent's own log file on the Hostinger VPS (/home/ima/logs/<agent>.log, where Hermes runs them), then cross-check the Agent_Log tab to see how far it got.
| Symptom | Likely cause → fix |
|---|
Does one AGENT.md make an agent run? — No, and here's what each folder still needs
The AGENT.md is the blueprint — it tells you (and Opencode, the model that writes the code) exactly what the agent does. A spec can't run itself. To actually run 24/7 each folder needs a worker plus the shared toolkit:
Inside each agent folder
AGENT.md (the spec — you have this) ＋ run.py (the worker — generated from the spec). That's it for the folder.
Shared, once, for all agents
The _shared/*.py toolkit (sheets, queue, gmail, opencode, gates, telegram, log), a filled .env, and Hermes firing each run.py on schedule.
So the plan is sound and your Hostinger VPS — managed by Hermes — is the right home; the one remaining build step is turning each runbook into its run.py.
Get One Week's of Work in One Day · Outdoor IMAs
The only agency built
for the outdoor audience.
Creators ready to deploy. Maximum effort is your duty — the outcome belongs to Allah alone.
—
Deep Work
0%
Get one week's of work in one day
Active Pitch Pipeline
0
Brands currently engaged
Roster Activation
0%
0 of 0 creators deployed
Check these off before moving to the next week.
Pipeline · Cold to Closed
Brand CRM
Every outdoor brand relationship — from cold ICP target to signed anchor partner. Pitch in disciplined rounds, protect margins, and keep one deal in focus at a time.
Income Generators
Onboard new brand
Title Case enforced · advancing to roster requires a confirmation gateBrand roster & pipeline
Roster · Source of Truth
Influencer CRM
Every vetted outdoor and travel creator, their platforms, reach, and live campaign stage. Raw, unviable and dropped leads are deliberately kept out of the working pipeline. The Roster Activation metric on the dashboard is calculated live from this list.
0
Total Influencers
$0
Profit (Talent Mgmt)
0
FIT
0
STORYFIT
🚨 Missed Talent Retainer Assets
There are currently 0 unrepresented outdoor creators in your wider network. At a conservative target commission of $1,000/month per creator, this represents a monthly uncaptured agency monetization gap of $0/mo.
$0
Monthly opportunity
0 managed · 0 in sheet
Pipeline Stages
Add a creator manually — set their handle, platform, country, niche, views and the stage they enter at. Every card has a ✕ remove control (with confirmation) so you can take anyone out at any stage.
Lead Generation Agent
Meet Atlas, your Lead Gen BOT.
Atlas builds targeted creator lead lists — by platform, niche angle, audience size, and country — using the agency's 15 sourcing methods. It finds and lists; outreach and closing stay with you.
0
Total Influencers
0 lists generated
0
Total Brands
0 lists generated
⚙️ Setup & integrations — deduplication & AI assistant
🔁 No duplicate leads
The uploaded master lists (All ENG / All EUR influencers, All ICP / Sub-niche brands, 2026 Q2) are loaded and de-duplicated. The influencer lead-gen now draws real, unique creator names from these lists and skips anyone already in the Influencer CRM; the brand vertical view surfaces real, unique brands from the lists and skips any already in the Brand CRM (all case-insensitive).
🤖 AI assistant (detailed feedback)
To have Atlas give richer, model-generated feedback, point this page at your Opencode/Hermes backend on the VPS (the same one the agents use) and let it return the assistant's reply. A model key can't be embedded safely in a public HTML file, so the live wiring goes through that small backend. Until then, Atlas runs on the built-in rules.
Live Command · The Battalion
Agents at Work
Nine specialist agents, one mission: keep every brand and creator moving toward closed, paid campaigns — without you in the loop, except at the gates.
System status
OPERATIONAL
—
Agents online
9/9
Actions today
0
Awaiting your nod
0
Follow-ups queued
0
Influencer Flank
Supply line — find, vet, sign creators
Brand Flank
Demand line — close brands, deliver, invoice
Shared Unit · Delivery & Invoicing
Where both flanks converge — drafts, go-live, money in
Live activity feed
Agent_LogFollow-Up Engine
8-step chase per stage · auto every 2–3 days