Claude-stack integrations · fixed price · no calls

AI integrations for your existing systems. Fixed price, working in 14 days.

I connect Claude to the tools you already run — your docs, your CRM, your inbox, your database — and hand over working automation with source code and a runbook. Three packages, prices listed below. You pick one, fill in a form, and I build. No sales calls, no negotiation, no scope creep.

3fixed-price packages
14 daysfrom confirmed scope to delivery
0sales calls in the process
01

Packages

Every package ends the same way: a working automation running against your systems, the full source repository, and documentation your team can operate without me. The price is the price.

Single Integration

$750fixed

One Claude integration wired into one system you already use.

Example: a RAG assistant over your company docs, answering questions inside Slack or your helpdesk.


  • One Claude-powered integration into one existing system
  • Document ingestion + retrieval (RAG) where the scope calls for it
  • Source repo with setup guide and .env template
  • Prompts and retrieval settings documented, not hard-coded
  • One round of in-scope fixes within 7 days of delivery

Multi-System Workflow

$1,850fixed

An n8n + Claude workflow spanning two to four systems, built to fail safely.

Example: inbound email → Claude classification → CRM update → notification, with a human approving before anything is written.


  • n8n workflow orchestrating Claude across 2–4 systems
  • Retry logic with backoff on every external call
  • Human-in-the-loop approval before irreversible actions
  • Failures routed to a channel or inbox you choose
  • Exported workflow JSON, source repo, and runbook
  • One round of in-scope fixes within 7 days of delivery

Full Pipeline Build

$3,500fixed

A complete pipeline, end to end: ingestion, processing, validated output, delivery.

Example: PDF-to-structured-data extraction feeding your database, or a CRM automation covering intake through follow-up.


  • End-to-end build: ingestion → Claude processing → validated structured output → destination system
  • Schema validation on every model output before it touches your data
  • Retries, dead-letter handling, and human sign-off on irreversible writes
  • Deployed on your infrastructure, with handover
  • Full source repo, runbook, and recorded walkthrough
  • Two rounds of in-scope fixes within 14 days of delivery

If your project doesn’t fit any package, I’ll say so after reading your intake form and we part ways cleanly — no upsell, no “custom quote.” Scope is confirmed in writing before any work or payment happens.

02

How it works

The whole process runs in writing. You never have to book a call, and I never ask you to.

01

Pick a package

The three packages and prices are listed above. That’s the whole price list — there is no negotiation step because there is nothing to negotiate.

02

Fill in the intake form

Ten questions about your systems, your data, and what “working” means to you. Fifteen minutes, roughly. The questions are listed below so you can prepare answers first.

03

Scope confirmed in writing

Within two business days you get a written scope document: what will be built, what is explicitly out, what access I need, and the acceptance check for day 14. Work starts only after you approve it. If your project doesn’t fit the package, I tell you that instead.

04

Build delivered in 14 days

Fourteen calendar days from approved scope: the working automation, the source repository, and the runbook. You run the acceptance check from the scope doc against the live build.

05

No calls, ever

Questions during the build are handled over email, in writing, with answers you can search later. If you want ongoing cover after handover, add Keep-Alive.

03

Demo builds you can read

This is a new practice, so instead of testimonials there are two public demo repositories. Each one is a working, documented build in the same style you’d receive — read the code and judge for yourself.

demo/claude-rag-crm-demo

view source on GitHub →

A retrieval-augmented assistant over a sample company document set. It answers questions with citations back to the source documents, and when a conversation surfaces a concrete request, it writes a structured, schema-validated summary onto the matching CRM contact record. This is the shape of the $750 package.

docs/ chunk + embed vector store claude-sonnet-5 schema check CRM write-back

What the repo shows

  • Ingestion script: chunking, embedding, and indexing a docs folder
  • Retrieval with source citations returned alongside every answer
  • CRM write-back that validates against a schema before any write
  • README covering setup, .env template, and cost notes per query

The Claude call at the center

# answer with citations, grounded in retrieved chunks
resp = client.messages.create(
    model="claude-sonnet-5",
    max_tokens=1024,
    system=CITE_SOURCES_PROMPT,
    messages=[{"role": "user",
               "content": build_prompt(query, chunks)}],
)

demo/n8n-claude-lead-pipeline

view source on GitHub →

An n8n workflow that takes inbound documents from a webhook, has Claude extract structured fields, validates the result, and then stops for a human. Nothing is written to the destination until a person clicks approve. Rejected or failed items land in a dead-letter path with the error attached. This is the shape of the $1,850 package.

webhook in claude-sonnet-5 extract validate (retry ×3, backoff) human approval write to destination/ dead-letter

What the repo shows

  • Exported n8n workflow JSON you can import and run locally
  • Claude extraction node with a strict JSON schema for output
  • Retry-with-backoff wrapper and a dead-letter branch for failures
  • Approval step: the workflow waits for an explicit human yes

Why the approval step matters

  • Model output never writes directly to systems of record
  • A person sees exactly what will be written, before it is
  • Rejections are logged with reasons, so prompts improve over time
  • The same pattern applies to CRM writes, refunds, and outbound email
04

Start the intake

Pick your package inside the form. You’ll get a written scope document within two business days — work and payment happen only after you approve it.

intake — by email

No form to fight with — the intake runs by email.

Copy the ten questions on the right into an email, answer them in plain language, and send. You’ll get a written scope document within two business days — work and payment happen only after you approve it.

hello@kelsonworks.com

The ten questions the form asks

  1. Which package? ($750 / $1,850 / $3,500 — and whether you want Keep-Alive at $350/mo)
  2. In one plain paragraph: what should this automation do?
  3. Which systems are involved? (CRM, email, docs store, database, chat, other)
  4. Where does the source data live, roughly how much is there, and in what formats?
  5. Where should the output go, and what action gets taken with it?
  6. Which steps need a human sign-off before they run?
  7. How will you grant access — API keys, a sandbox, or a test account?
  8. What check will you run on day 14 to call this “working”?
  9. Hard constraints: data that can’t leave certain regions, tools I must or must not use.
  10. The email address for the written scope document.

Send your answers to these ten questions to hello@kelsonworks.com — same process, same two-business-day scope turnaround.