How we engagePricingCase studiesBlog
How we engage

Under the hood.

An engagement is less about what we deliver on week one than what keeps running after week twelve. The pages below trace every claim on the home page to a real component in a real repo.

The path

From the first call to your fleet shipping.

Discovery

Is this the right fit?

A 30-minute conversation, free, no commitment. We learn how your business operates, where the bottleneck is, and whether agents are the right tool. If they're not, we say so. You leave with a clear picture and no follow-up sequence.

Pilot

One agent in your repo.

A $2,500 to $4,000 fixed-scope build that ships one working agent in five working days (ten for Custom). SEO, Content, Outbound, Support triage, or custom (picked at scoping). You see what the fleet feels like before committing to running one.

Pilot fee credits 100% towards a Full Engagement if you continue within 30 days.

Full Engagement

The longer engagement.

  • ✦Operate: runs your fleet for you.
  • ✦Build: ships product or engineering work with the fleet.
  • ✦Transfer: installs the architecture, trains your team, and hands you the keys.
§ 0  ·  Who this fits

Your own time is your biggest constraint.

The bottleneck isn't a missing tool or a missing process. It's that every meaningful piece of work: the blog post, the outbound email, the support reply, the bug fix, the half-built admin tool: needs your judgment before it ships. The fleet is built so judgment is the only thing that needs your time.

Founders, function leads inside a larger company, or anyone with authority to change how the work gets done. Operate covers the running. Build covers the shipping. Transfer covers the handoff. Pick the shape that matches what you want at the end.

Common fit signals
You're the bottleneck for content, SEO, outbound, support, or internal tooling: and hiring hasn't fixed it.
You've tried 'using more AI' and hit the mental wall, not the tooling wall. The models are fine; you don't yet have the operating discipline to deploy them.
You want your team to own the system at the end. Renting access to someone else's agents isn't the outcome you're looking for.
You have a codebase the agents can live in, or you are open to us standing one up for you. Stack-agnostic: Next.js, Rails, Django, Go, or a fresh repo we set up at engagement kickoff.
You'd rather review than create.
§ I  ·  The architecture

Your repo is the coordination layer.

YOUR_REPO/apps/├ web/├ api/agent-reports/└ admin/marketing/docs/agents/├ state/ ← daily state files├ prompts/ ← constraints live here└ briefs/ ← agent outputs.git/ ← every commit bylinedMARKETING SQUADRONagent.contentagent.seoagent.salesagent.socialagent.adsagent.supportPRODUCT SQUADRONagent.designagent.frontendagent.backendCMO FLAGSHIPreads 6 marketing briefs→ 1 summary / day→ flags what needs youCTO FLAGSHIPreads 3 product briefs→ PR reviews & scope→ constraints violationsFOUNDERone messageeach morningthree thingsthat need you
I.
Git as coordination
No message queue, no event bus. Every agent reads & writes files in your repo. You can grep the last 90 days of every marketing decision.
II.
Reports, not telemetry
Agents POST daily briefs to /api/agent-reports in your app. Data lives in your Postgres. Admin dashboard at /admin/marketing shows the last 7 days per agent.
III.
Two synthesizers
CMO reads 6 marketing briefs; CTO reads 3 product briefs. Reading 9 daily briefs is more work than you'll do. They read for you.
IV.
Constraints, not autonomy
Each prompt includes hard limits: which files it can touch, which commands it can run, what cannot deploy without your approval.
§ II  ·  A day in the fleet

Twenty-four hours, end to end.

06:00
Agents wake
Scheduled via cron on your VM. Each agent pulls latest main, reads its prompt and state file, begins its shift.
all agents
08:15
Content drafts land
agent.content commits two draft posts to a branch. Bylined: Authored-by: agent.content <content@armada.your-company.com>.
agent.content
09:40
SEO sweep
agent.seo re-crawls the site, opens a PR rewriting 14 meta descriptions and 3 H1s. Tagged for review.
agent.seo
11:00
Sales outreach
agent.sales sends 27 personalized first-touch emails. Each tied back to the prospect record in your CRM.
agent.sales
14:30
Product PR
agent.frontend finishes a component refactor. agent.design has already reviewed the Storybook diffs. CTO flagship queues the PR for your look.
agent.frontend + agent.design
16:00
Support triage
agent.support has triaged 41 tickets: 38 auto-resolved, 3 escalated to you with full context and suggested replies.
agent.support
18:00
Daily briefs fire
Every agent POSTs a structured brief: what shipped, what blocked, what it flagged. Data lands in your Postgres.
/api/agent-reports
21:00
Flagships synthesize
CMO reads 6 marketing briefs. CTO reads 3 product briefs. Each writes a 200-word summary, flagging only what needs your judgment.
CMO + CTO
06:30 +1d
Your inbox
One message. Three things. No dashboards to check unless you want to: everything is auditable, nothing is required.
you
§ III  ·  The constraint

The best agents are the most constrained ones.

Every agent prompt includes three sections: its role and voice, its daily cadence, and: most important: its limits. Which files it can touch. Which commands it can run. What it must never ship without your approval.

Here's an abridged version of one we actually ship.

docs/agents/prompts/content.md- abridged
# role
You are agent.content, a staff writer for $COMPANY_NAME.
You write like the founder, not like a content marketer.

# cadence
Ship one long-form post per week, two shorter posts per week.
Always leave work on a feature branch. Never push to main.

# limits (hard)
allow_write:
  - content/posts/**
  - docs/agents/state/content.md
deny_write:
  - package.json, infra/**, .env*
require_review:
  - any post tagged `customer-story`
  - any claim involving revenue or headcount

# reviewer
Before committing, run your draft past agent.reviewer.
If it flags slop, revise. If it flags twice, escalate to CMO.
█
§ IV  ·  Day one

Three things to have ready.

Everything runs on infrastructure you own and pay for directly.
I.
An Anthropic account
~$200/mo for a typical fleet
You pay Anthropic directly. We never touch billing and you can cap usage at the account level. Token spend shows up in your own dashboard.
II.
A small cloud VM
~$10/mo · your DigitalOcean, Fly, or AWS
Agents run as scheduled Claude Code sessions on your box. We SSH in to deploy and maintain; you can rotate the key the moment we leave.
III.
GitHub collaborator access
your repo, your branch protection
We add two handles as collaborators. Every commit from an agent is bylined. Every human commit from us is reviewed like any other contributor.
We're an SSH key away while we're engaged: and a key rotation away when we're not.
§ V  ·  Stack

Will it work for my stack?

The agents themselves work in any codebase, and we can stand one up if you do not have one. They are Claude Code sessions running against your repo, language-agnostic.

What is stack-specific is the reporting layer: a POST endpoint, a dashboard, a database table. Our reference implementation is Next.js + Supabase. Other stacks need the endpoint and dashboard re-implemented: roughly a day, which one of our agents does as part of onboarding.

Next.js + Supabase
0 hours
Our reference stack. Zero reimplementation needed.
Next.js + Postgres
~2 hours
Supabase client swapped for direct Postgres. Straightforward.
Rails / Django
~1 day
Reports endpoint rebuilt in your framework. Dashboard is a simple admin view.
Go / Elixir / Phoenix
~1 day
Same story. Agents don't care what the endpoint runs.
No web app at all
~2 days
We stand up a small reports service on your VM and a fresh repo for the agents. Suitable for non-technical businesses (event planners, agencies, professional services) where marketing is the bottleneck and there is no existing codebase to extend.
§ VI  ·  Questions founders ask first

Before we get on a call.

How is this different from hiring an agency?+ read

Agencies produce deliverables. We produce a system that produces deliverables, which is yours. When our engagement ends, an agency relationship ends with nothing in your hands. Ours ends with a working system committed to your git history.

How is this different from SaaS tools like Clay, Copy.ai, or Jasper?+ read

Those are tools your team has to operate. Armada is the team. We deploy the tools (our own and others), run the workflows, and report outcomes. You get the output, not another subscription to manage.

What if I don't want to run the system myself at the end?+ read

That's the Operate tier: we keep running it. Transfer is optional. Nothing forces you to take the keys.

Can agents touch production?+ read

Only with explicit per-agent grants. The default posture is: agents commit to branches, humans approve merges to main, deployments require founder approval. You can loosen this per-agent (e.g. let agent.seo deploy meta changes) but we don't start there.

What happens when an agent makes a mistake?+ read

Because every action is a commit, you can revert. Because every agent has a byline, you can retrain or disable the one at fault. Because reviewer agents sit between work and main, most mistakes never reach main at all.

How fast can we start?+ read

A Pilot starts within a few days of the discovery call once repo access is provisioned and ships in 5 working days (10 for Custom). Operate or Build can start within a week once access is provisioned. Transfer is a 2 to 4 week engagement end-to-end. We won't onboard you if we don't think the fit is real.

Want this on paper?
The Agent Readiness Guide.

A 13-page PDF that condenses the architecture, the engagement options, and the readiness checklist into a printable take-home version. Same content, different format. Free, lands in your inbox.

Get the agent readiness guide  →
◆   Thirty minutes. No pitch deck.   ◆

Still have questions?

Bring your weirdest one. We'll either answer it, or tell you who can.

Book the call →