White-label AI support agents for software companies.
Turnkey SaaS. Multi-LLM. Production-proven.
The Problem
of support tickets are repetitive, answerable from existing docs
average cost of a single human support interaction
to build a custom AI support system with ML engineers
SMBs can't afford dedicated support teams or AI engineers. They have the documentation but no way to turn it into an intelligent, always-on assistant.
The Solution
A turnkey B2B SaaS platform that converts any documentation set into a production-grade AI chat agent — white-labeled under your customer's brand, with zero ML expertise required.
MD, PDF, Wiki exports
12 branding fields + CSS
Grok, Claude, GPT-4o
Proven 4-hr deploy
How It Works
Customer provides Markdown, PDFs, or wiki exports. KBForge chunks, structures, and builds a FAISS vector index automatically.
Name, logo, colors, custom domain, system prompt. 12 white-label config fields — their users see their brand, not ours.
Grok-mini (cheapest), Claude Sonnet (smartest), or GPT-4o. Customers can switch models at any time.
Isolated Docker container, auto-SSL via Caddy, streaming chat, user auth, admin analytics, feedback system — all included.
Proof point: The SEM Textbook Agent was deployed from zero to production in ~4 hours using a Copilot agent session — different domain, fully white-labeled.
Market Opportunity
Total Addressable Market
Global AI customer support & knowledge mgmt
Serviceable Addressable
SMB software companies with documentation
Serviceable Obtainable
Realistic year-3 capture
LLM costs dropped 90%+ in 18 months — now viable for SMB SaaS
RAG architecture is mature — FAISS + embeddings deliver accurate, grounded answers
Every software company has docs but can't afford ML engineers
Support costs rising while customer expectations for instant answers grow
Business Model
Solo devs, small OSS
SMB software companies
Compliance & scale needs
Traction
Tests passing
Domains deployed
First KB buildout signed
New-domain deploy time
Complex scientific software (electron microprobe analysis). 2 knowledge bases, multi-LLM support. Production-deployed serving real users.
Signed pilot: $10,000 KB buildout + $200/mo maintenance. Validates both the product offering and the professional services revenue stream.
Competitive Advantage
Not a prototype — 639 tests, multi-KB, streaming, auth, analytics, feedback. Ship on day one.
12 branding fields, custom domains, custom system prompts. Customers' users never see "KBForge."
Not locked to one provider. Grok (cheapest), Claude (smartest), GPT-4o — swap in seconds.
4-hour proof point (SEM Agent). Competitors take weeks or months of custom engineering.
~$100–$200/mo fixed cost. Shared VPS fits 4–8 tenants. 70%+ gross margins from day one.
KB buildout services ($1.5K–$15K) land customers who convert to recurring SaaS revenue.
The Team
CEO / Founder
Bio — background, relevant experience, domain expertise
Role — e.g., CTO
Bio — engineering background, relevant skills
Role — e.g., Head of Sales
Bio — go-to-market experience
Hiring plan: 4–5 people in year one — engineering, sales, and customer success.
The Ask
MVP launch + first 5 paying tenants
Self-serve onboarding, Stripe billing live
20 tenants • $__K MRR
Onboarding wizard, multi-KB support
50 tenants • $__K MRR
Enterprise tier, SSO/SAML, API export
Seed-ready • $__K MRR
Position for $2–3M seed round
hello@kbforge.com • kbforge.com
Appendix
Prepared answers for investor due-diligence and prospective customer objections.
Investor Q&A
Those are support-desk platforms that bolted on AI. We're the opposite: an AI-first agent that's fully white-labeled and deployed in hours, not months. No CRM lock-in, no per-seat pricing. Customers get a standalone, branded AI expert — not a chatbot tab inside someone else's ticketing system. We also support multiple LLM providers (Grok, Claude, GPT-4o) while incumbents are locked to one.
Raw ChatGPT has no knowledge of your documentation — it hallucinates. OpenAI's Assistants API requires engineering effort (chunking, embedding, prompt engineering, auth, analytics, feedback). KBForge does all of that out of the box: FAISS vector retrieval, tiered search, source citations, user auth, admin analytics, and white-label branding. It's the difference between a raw API and a finished product.
LLM providers will always optimize for horizontal, generic use cases. Our moat is vertical depth: production-grade white-label (12 branding fields, custom domains, system prompts), multi-tenant isolation, per-tenant analytics, feedback loops, and the professional services flywheel that lands customers. We're also LLM-agnostic — if OpenAI ships RAG, we can use it as a backend while still owning the customer relationship and brand layer.
A feature would be "add AI to your existing help desk." KBForge is a full platform: ingestion pipeline, vector indexing, multi-LLM orchestration, per-tenant containers, billing, usage metering, admin dashboards, user auth, feedback capture, and white-label deployment. That's 639 tests worth of product. The services business (KB buildout at $1.5K–$15K) creates a land-and-expand motion that a feature can't replicate.
There are ~30 million small businesses globally, and millions of software companies specifically. Most SMB software companies rely on docs + email support. They can't afford Zendesk Enterprise or a support team, but they do have documentation. That's our sweet spot: the underserved middle between "DIY wiki" and "$50K/yr support platform."
Investor Q&A
Starter is a land tier. At $149/mo with ~$6–$12/mo infra cost and ~$20–$40/mo LLM cost (2K queries), gross margin is ~50–60%. But Starter customers who outgrow 50 users or 2K queries upgrade to Pro ($399/mo, 70%+ margin). The real economics are in Pro and Enterprise. Starter also drives word-of-mouth from OSS/indie developers — low-cost acquisition channel.
Early CAC is near-zero (founder-led sales, content marketing, community). At scale, target CAC is $500–$1,500. Pro customers at $399/mo with 18-month average retention = ~$7,200 LTV. That's a 5–14x LTV:CAC ratio. Professional services ($1.5K–$15K KB buildouts) are CAC-negative — we get paid to acquire customers.
Year 1: likely 50/50 as KB buildouts land customers. Year 2+: target 80% SaaS / 20% services as self-serve onboarding matures. Services don't need to scale infinitely — they're a customer acquisition tool. Over time, we'll productize the ingestion pipeline so more customers self-serve, shifting the mix toward recurring revenue.
LLM costs have dropped 90%+ in 18 months and the trend continues. But we're hedged: multi-LLM means we can route to the cheapest provider. Our overage pricing ($0.02–$0.05/query) gives 2–5x margin over LLM cost even at today's prices. If costs spike, we adjust pricing tiers — standard SaaS practice.
They could, just like any SaaS customer could rebuild in-house. But they'd need to maintain the RAG pipeline, manage LLM API keys, handle auth/security updates, monitor uptime, and update the KB as their docs evolve. Our $200–$1,500/mo is cheaper than an engineer's time. Plus, the ongoing KB maintenance retainer creates switching costs.
Investor Q&A
The $10K KB buildout is services, but it converts into a $200/mo SaaS subscription — that's the model working as designed. The pilot also validates willingness to pay for both components. We're building the self-serve pipeline in parallel so future customers can onboard without custom work.
We're pre-launch — the pilot is validation, not the pipeline. With funding, we'll launch the marketing site (already built), attend developer conferences, do content marketing around "AI support for docs," and leverage the CalcZAF + SEM case studies as proof points. Target: 5 paying tenants in Q1 post-funding.
We proved generalization with the SEM Textbook Agent — completely different domain, deployed in 4 hours. The architecture is domain-agnostic by design: 12 white-label config fields, pluggable knowledge bases, and a system prompt that adapts to any subject matter. The CalcZAF domain was the hardest test case (niche scientific software); consumer-facing docs are easier.
Container density — currently Docker Compose, which works up to ~50–100 tenants per node. Beyond that, we migrate to Kubernetes. The migration path is clear (containerized services, stateless app, external volume mounts), and K8s orchestration is a solved problem. We've budgeted this for the Q3 milestone.
We built the entire product — 639-test codebase, production deployment, multi-domain proof — before raising a dollar. That's execution speed most startups can't match. Domain expertise in RAG architecture, LLM orchestration, and the SMB support workflow means we're shipping product, not learning the space.
55% of funds ($275K) goes to engineering — enough for 2–3 hires at competitive startup salaries for 12 months. We're targeting strong mid-level engineers who want early equity + startup upside. Remote-first keeps costs lower than SF/NYC. The founder continues full-time product work, so we're effectively a 3–4 person engineering team from day one.
Client Q&A
Each tenant gets an isolated Docker container with its own file system, database, and API keys. No data is shared between tenants. Documentation is stored on our infrastructure (DigitalOcean/Hetzner, US or EU) and accessed only by your container. We never use your data to train models — it's processed through LLM APIs that also don't train on API inputs (OpenAI, Anthropic, and xAI all confirm this in their API terms).
Conversations are stored in your tenant's isolated database for analytics and feedback review. You have full control — admin dashboard lets you view, export, or delete conversations. On the Enterprise plan, you can configure retention policies (auto-delete after N days). If you cancel, we archive and delete all data within 30 days (or immediately on request).
No. We use LLM APIs (not training endpoints). OpenAI, Anthropic, and xAI all confirm that API inputs are not used for model training. Your documentation stays in your container, your conversations stay in your database. We will provide a DPA on request for Enterprise customers.
GDPR: we're compliant by design (data isolation, deletion on request, EU hosting option). SOC 2: on our roadmap for Q3 — required for Enterprise tier. HIPAA: not currently supported, but the isolated-container architecture makes a BAA feasible for a future healthcare vertical. Enterprise customers on dedicated VPS can bring their own compliance requirements.
ChatGPT doesn't know your docs — it can only guess based on training data. KBForge builds a FAISS vector index of your specific documentation and retrieves relevant sections before generating each answer. Every response includes source citations so users can verify. Plus: your branding, your domain, user authentication, admin analytics, feedback capture, and no risk of the AI discussing competitor products or off-topic content.
Three layers: (1) Tiered retrieval ensures the LLM only sees relevant documentation chunks, not the entire internet. (2) System prompts instruct the model to say "I don't have information about that" when retrieval confidence is low. (3) Source citations on every answer let users verify, and the feedback system lets them flag bad answers for admin review. This is fundamentally different from raw ChatGPT, which has no retrieval grounding.
Users can click a feedback button (thumbs down + optional comment, with screenshot capture). Admins see all flagged responses in the analytics dashboard. You can then update KB articles to improve future answers, adjust the system prompt, or add the question to a "known issues" list. The feedback loop continuously improves accuracy.
Client Q&A
Yes. Our ingestion pipeline handles Markdown, PDFs, and wiki exports (Confluence, GitBook, ReadTheDocs). For messy sources, our professional services team ($1.5K–$3K) will clean, structure, and chunk your content. For source-code-heavy projects, we offer agent-assisted KB authoring ($5K–$15K).
Upload updated docs anytime — we rebuild the FAISS vector index automatically. On our maintenance retainer ($500–$1,500/mo), we handle ongoing updates as your software and docs evolve. Self-serve customers can re-trigger the indexing pipeline via admin panel.
Yes — this is a core design principle. When FAISS retrieval finds no relevant documentation, the system prompt instructs the LLM to respond with "I don't have information about that in my knowledge base" rather than guessing. You can customize this fallback message.
Yes. The custom system prompt (Pro and Enterprise plans) lets you define boundaries: "Only answer questions about [product]. Do not discuss competitors, pricing, or legal matters." The agent inherently stays within your KB scope since retrieval is grounded in your documentation.
No. Full white-label: your company name, your logo, your colors, your domain. 12 configurable branding fields plus full CSS control. Your users see "Acme Support Agent" at acme-support.com — no mention of KBForge anywhere in the UI.
Currently deployed as a standalone web app on your custom domain. Embeddable widget (iframe/JS snippet) is on our roadmap. Enterprise customers can request iframe integration as a custom feature today.
Not yet natively. Enterprise plan includes API export of conversations and feedback. Webhook-based integrations (escalate to Jira ticket when confidence is low) are on the Q2 roadmap. We prioritize integrations based on customer demand.
Pro: 99.5% uptime target. Enterprise: 99.9% SLA with dedicated VPS and uptime monitoring (UptimeRobot + Sentry). If our service is down, the support page shows a graceful fallback message — it doesn't break your site.
We can set a hard query cap (agent stops responding) or a soft cap (alerts you, keeps responding). Enterprise plans include custom overage terms. You always see real-time usage in the admin dashboard so there are no surprises.
Yes. Free 30-day pilot for early customers — we set up your agent and help load your knowledge base at no cost. After 30 days, pick a plan or walk away. No credit card required to start.
The underlying LLMs (Claude, GPT-4o, Grok) support 50+ languages natively. If your docs are in English, the agent can still answer in the user's language — the LLM handles translation on the fly. For best results, provide KB content in the primary language(s) your customers use.
Yes. The admin analytics dashboard shows: total conversations, popular topics, unanswered questions, feedback scores, and flagged responses. Full plan includes analytics; Enterprise adds API export. This data lets you continuously improve your KB and identify gaps in your documentation.