AgentForge · Custom AI Assistants
A custom AI assistant for your shop, your home or your hobby.
Don't know how to build a bot? Tell us what you need and we'll build it. Every delivery ships with test cases.
02 · The problem
Generalist AI. Specialist work.
Most AI assistants fail quietly: they make things up, drift off-brand and skip the rules. Off-the-shelf AI is a generalist. Your shop, your home or your hobby needs one that knows your details, your tone and your limits.
We write the instructions that address that, then check them against a written test suite. You get a tested system prompt, an assistant installed in your own AI account, or a full agent blueprint. Fixed price, fixed scope, any model.
03 · Tiers
Three fixed-scope builds.
Pricing is competitive and will save you time and money. You know the price, the scope and the deliverables before you pay.
One AI job done rightPrompt Sprint
One production-grade system prompt, built and tested.
- 1 system prompt covering role, rules, tone, output format, refusal and escalation behavior
- Guardrails against hallucination and prompt injection
- A 10-case test suite with pass/fail results
- A usage guide and a change log
- Works in ChatGPT, Claude, Gemini, Grok or any API
You receiveA 1-page quick-start, the system prompt, a skill version, setup guides for ChatGPT, Claude, Gemini and Grok, sample prompts, a usage guide, 10-case test results and a change log.
Turnaround: 3 business days2 revision rounds30-day revision window
A team assistant that knows your businessAssistant Build
A ChatGPT Project, Claude Project or Skill, Gemini skill, or Grok Project, configured and installed in your account.
- Everything in Prompt Sprint
- Knowledge files audited (up to 10 docs)
- Up to 2 connected actions
- Conversation starters and a model/settings recommendation
- 25-case test results
- Installed in your account (you grant access), or you self-install with the 1-page quick-start
- A 30-minute handover call and an admin guide
Plan you'll needA plan that supports projects or skills on your platform. The assistant runs in your account, on your plan. A Free ChatGPT project holds 5 files, so a 10-doc knowledge base needs a paid plan.
Turnaround: 5 business days3 revision rounds30-day revision window
Multi-step workflows across toolsAgent Blueprint
A plan for up to 4 connected bots, ready to build.
- A plan for up to 4 connected bots
- A prompt for each bot, plus routing logic
- All tool schemas
- Architecture and running-cost estimate (stack, e.g., OpenAI Agents SDK, LangGraph, n8n)
- A 40-case test plan
- A 60-minute walkthrough
- 14 days of Q&A after delivery
- Building it
- The Blueprint is the spec. Implementation is quoted separately, based on the build roadmap, or your developer can build from it.
- After the 14 days
- Q&A ends. Revision rounds stay open for the 30-day window. After that, use the monthly tune-up or ask for a quote.
- Scheduled or always-on
- Need your bot running 24/7 or on a website? We start with a Blueprint, then send a fixed quote. Any monthly fee is listed in your quote before you pay.
Turnaround: 10 business days3 revision rounds + 14 days of Q&A30-day revision window
Add-ons
RushFaster delivery; rush also applies to revisions.
Extra botOne more bot with its own prompt, tests and setup guide.
Extra promptOne more system prompt for a separate job, tested.
Monthly tune-upOne revision round a month plus a re-test after major model updates. Starts after the 30-day revision window. Unused rounds don't roll over.
Scheduling setupWe set your bot to run on a schedule inside your own ChatGPT tasks, Grok Automations or Claude scheduled tasks, with one test run and a "pause, change or delete" guide.
Always-onNeed your bot running 24/7 or on a website? We start with a Blueprint, then send a fixed quote. Any monthly fee is listed in your quote before you pay.
You get a Bot Package installed in your own account. A shared link is a demo only. You own the delivered files.
Revisions
Each package includes revision rounds for 30 days after delivery (2 for Prompt Sprint, 3 for Assistant Build and Agent Blueprint). Send one consolidated change list to [email protected] with the subject 'Revision – order # – bot name'. We turn each round around in 2 business days, or 1 business day with rush, re-test, and send an updated version with only the changed files. New jobs, platforms, extra bots or prompts, more than 2 actions or 10 docs, and requests after 30 days are quoted as add-ons or a new build.
04 · Compare
Side by side.
The same three tiers, line by line.
| Feature | Prompt Sprint | Assistant Build | Agent Blueprint |
| Best for | One AI job done right | A team assistant that knows your business | Multi-step workflows across tools |
| What you get | A tested system prompt | An assistant installed in your account | A plan for up to 4 connected bots |
| Where it runs | ChatGPT, Claude, Gemini, Grok or any API | ChatGPT Project, Claude Project or Skill, Gemini skill, or Grok Project | The stack you choose (e.g., OpenAI Agents SDK, LangGraph, n8n) |
| Testing | 10-case test results | 25-case test results | 40-case test plan |
| Knowledge docs | Your own files, not audited | Audited and restructured, up to 10 | As scoped |
| Actions and tools | — | Up to 2 connected actions | All tool schemas |
| Quick-start, skill version, sample prompts | Yes | Yes | Yes, per bot |
|---|
| Setup | Paste-in guides for 4 platforms | Installed in your account, or self-install | Setup guides for any chat-platform bots |
|---|
| Architecture | — | — | Architecture and running-cost estimate |
| Call | — | 30-minute handover | 60-minute walkthrough |
| Turnaround | 3 business days | 5 business days | 10 business days |
| Revision rounds | 2 | 3 | 3 |
| Revision window | 30 days | 30 days | 30 days |
| Guides and support | One-page usage guide | Full admin guide | 14 days of Q&A |
Sample deliverableSee a redacted Agent Blueprint.
A complete sample for a fictional plumbing company: architecture, prompts, tool schemas and test cases.
05 · Why AgentForge
Tested, not vibes.
Tested, not vibes
Every delivery ships with test cases and pass/fail results.
Model-agnostic
Works on ChatGPT, Claude, Gemini, Grok or the API.
You own it
You own the delivered files. Plain-English docs, no lock-in.
Fixed scope
You know the price, the date and the deliverables before you pay.
06 · How it works
Five steps.
- Complete the short intake below.
- We reply by email and confirm the tier, platform, plan and price. You reply "Approved".
- You get a Stripe invoice once you approve your quote. When it's paid, we build and test on your platform and plan.
- We install it in your account, or you self-install with the quick-start.
- You request revisions within 30 days; we refine.
Payment is by Stripe invoice. You'll get an invoice once you approve your quote; work starts when it's paid. Submitting the intake costs nothing and commits you to nothing.
07 · FAQ
Questions.
Which AI platform do I need?
ChatGPT, Claude, Gemini or Grok. Your bot is installed in your own account as a ChatGPT Project, Claude Project or Skill, Gemini skill, or Grok Project, and runs on your plan. Tell us your platform and plan in the intake.
Do you build the bot code?
The Blueprint is the spec. Implementation is quoted separately, or your developer can build from it.
Is my data safe?
Share only what the assistant needs. We work in your account and will sign an NDA on request. We keep only the delivered package files, not your data, for the revision window.
What if it doesn't work?
Every tier includes revision rounds tied to the test suite. Failing test cases are what revisions are for.
All questions →
ContactQuestions, revisions or an order already underway? Email us.
08 · Start the intake
Tell us what you need.
A few questions, a couple of minutes. We'll reply by email with next steps.