Open Source · Apache 2.0 · BYOK
The AI search bot you were going to pay $99/month for.
We got tired of chatbot SaaS companies charging rent on your own API keys. So we built the whole thing, open-sourced it, and let you run it on your keys, your database, your rules. Upload your content, embed one script tag, done.
Start FreeGitHub coming soon
✓ No credit card ✓ No telemetry ✓ No token markup ✓ No lock-in
~$1200 convos / month
~9 KBWidget size
$0Self-host floor
Do the math$99↑ price increasing
cost_comparison.liveBYOK
200 support conversations / month
GB
GenericBot AI
CC
CommonChat · hosted
$0.98in API costs
Self-hosted$0.00
The AI costs half a cent per conversationThey charge $99 for a dollar's worth of tokensPriced at cost, because spite is a business modelZero token markup, everThe AI costs half a cent per conversationThey charge $99 for a dollar's worth of tokensPriced at cost, because spite is a business modelZero token markup, ever
The math they hope you never do
The AI is the cheap part. It always was.
A typical support conversation runs about 2,500 tokens in and 500 out. On Claude Haiku, at published prices, that is roughly half a cent. Here is the full breakdown for 200 conversations a month, and what everyone above that dollar is actually charging you for.
cost_breakdown · 200 convos
Input500,000 tokens @ $1 / million$0.50
Output100,000 tokens @ $5 / million$0.50
Raw AI cost / month~$1 /mo
One dollar. That is the raw material cost of "AI-powered support" at small-business scale. A $99 plan rents you a dollar's worth of AI at a 100x markup.
Brand X
CommonChat
Pricing philosophy
Expensive, and raises prices once you depend on it
Spitefully cheap, and structurally unable to trap you
200 convos / month
$30–$200 / month
~$1 in API; free to self-host
What happens to the AI cost
Marked up per message, hidden in tiers
Passed through at cost. We never touch it
When they disappoint you
You're stuck
Self-host free and keep everything
Setup & cost
Live in ten minutes. Paying almost nothing.
Other platforms mark up every token and call it "usage-based pricing." We call it a middleman fee. CommonChat runs on your own keys, so you pay Anthropic and OpenAI directly at list price, and nobody else.
Zero to a working bot in five steps
An onboarding wizard that even drafts your bot's personality and answer format for you. Editable, with a Regenerate button, because AI first drafts are AI first drafts.
Pay pennies, not platform fees
Bring-your-own-key pricing. A Haiku or gpt-4o-mini bot costs most sites less per month than a fancy coffee. We never touch, proxy, or mark up your inference.
Embed with one line of code
A single script tag with your bot ID. No SDK, no build step, no "talk to sales."
Skip vendor lock-in entirely
A plain Postgres + pgvector stack. No proprietary vector database, no separate search service, no hostage negotiations when you want to leave.
embed.html
<!-- paste once, anywhere -->
<script src="chat.commonpeople.co/w.js" data-bot="cc_9f2"></script>
● Hi! Ask me anything about our docs
One-line script tag, then the bubble pops onto the page
Branding
Your bot. Your colors. Our name nowhere near it.
Most chat widgets are a billboard for the chat widget company. Yours should look like it belongs on your site, because it does.
Match your site pixel-for-pixel
A four-color palette, font family, and border radius controls. Header, user bubble, send button, and launcher icon, each set independently.
Give the bot an actual personality
A custom system prompt up to 20,000 characters. Formal, funny, unhinged, your call. You can even override the answer format entirely.
Start conversations before visitors type
A custom welcome message and up to eight tappable starter questions.
Preview before your visitors see it
A live test chat in the dashboard. Break it in private, ship it in public.
theme_preview
Same widget cycling through brand themes live
Knowledge & search
It answers from YOUR content. Not vibes.
Generic chatbots confidently make things up. CommonChat is a search bot: it retrieves from what you gave it, answers in a ranked list with source links, and by default is forbidden from freelancing.
Feed it almost anything
PDF, DOCX, XLSX, CSV, TXT, Markdown, and HTML, each with a parser that understands the format instead of shoving raw bytes at an LLM.
Index your whole site without copy-pasting
Point it at a URL or sitemap, or bulk-load up to 1,000 pages from a single CSV (and export them back out for editing).
Refresh without re-paying to embed
Every page carries a content hash, so re-crawls skip anything unchanged and prune anything gone. Your embedding bill stays boring.
Keep the bot honest
Retrieval-only mode (default on). Every answer is a "Here's what I found" list with titles, descriptions, and source links.
knowledge_base
report.pdfdocs.docxdata.xlsxfaq.mdsitemap.xml
Knowledge base
Pending → Ready
248
File types funnel in, status chips flip to ready
Flexible embeds
One widget. Three moods. Nine kilobytes.
Some chat widgets ship half a megabyte of JavaScript to render a button. Ours is ~9 KB, held under a CI-enforced budget, and it doesn't load the chat app until someone actually opens it.
Keep your page fast
The chat iframe lazy-loads on first open, so visitors who never chat never pay for it.
Lock it to your domains
A per-bot allowlist of up to 100 origins, enforced at the API and in the CSP. Someone hotlinks your bot? It just doesn't work. Tragic.
Stop worrying about CSS fights
A sandboxed iframe with a locked-down postMessage bridge. Your theme can't break the chat and the chat can't break your theme.
Follow a recipe for your platform
Step-by-step guides for Squarespace, WordPress, and Webflow, plus a full embed API reference.
origin_check
GET yourdomain.com/chat → 200 OK ✓
GET stolen-embed.biz/chat → 403 Forbidden ✕
// allowlist enforced at API + CSP
// bundle: 9 KB · under CI budget
Works on your domain, 403s on a stolen embed
Usage tracking
See exactly what it costs. Because it's your bill.
Platforms that resell tokens have every incentive to hide usage. We don't sell tokens, so we just show you everything.
Track messages and tokens per bot
Per model, over any date range, with estimated dollar costs computed from published list prices.
Switch models without losing the plot
The model is recorded on every message, so cost estimates stay accurate even after you swap Sonnet to Haiku mid-month.
Cap your own spend
Per-bot rate limits you control: messages per minute and per day, enforced server-side. Your bot can't get hugged to death.
Numbers nobody else is harvesting
Zero telemetry, zero analytics pixels, zero phone-home. Every figure computed from your own stored data.
/usage
| Bot | Model | Msgs | Tokens | Cost |
|---|---|---|---|---|
| Support | Haiku | 412 | 1.2M | $1.10 |
| Docs | gpt-4o-mini | 188 | 0.6M | $0.34 |
| Sales | Sonnet | 63 | 0.3M | $0.92 |
network inspector · outbound telemetry:
// none
Real per-bot costs. Outbound trackers: none
Chat analysis
Your visitors are telling you what they want. Read it.
Every conversation is a free focus group. Most platforms bury transcripts or charge extra for "insights." We think reading your own data shouldn't be an upsell.
Read every conversation
A browsable, paginated transcript view with citations, token counts, model, and latency per message.
An AI-written report in one click
Top questions with counts, trouble points paired with concrete fixes, and what's working, analyzed on your own key and targeted to your site.
Watch trends over time
Every insights report saved to history, so you can compare what people asked in March against what they ask now.
Turn chats into contacts
Optional lead capture: up to 20 custom fields, shown before the first message, exportable as CSV or JSON. It's your data. Wild concept.
insights_report
// generating report…
✓ Top question: "Where's my order?" ×142
✓ Trouble point: refund policy hard to find
→ Fix: add a refund entry to the KB
✓ Working: shipping answers rated helpful
One click assembles the report, section by section
Teams & trust
Built for teams. Not priced per seat like it's 2019.
Invite your people, give them the right role, and stop sharing one login in a password manager like criminals. Then trust the boring security work underneath it.
Organize by team, not by hack
Organizations, owner/admin/member roles, and an org switcher for people who wear multiple hats.
Skip passwords entirely
Magic-link sign-in plus optional Google OAuth. Nothing to leak, nothing to reset. Invitations are expiring and token-based.
Keep API keys actually secret
Envelope encryption (AES-256-GCM, per-row keys). Never logged, never sent to the browser, displayed as ••••1234 forever.
Trust the boring security work
Authorization re-checked on every route, security review on sensitive paths, secret scanning on every PR. It's open source. Go audit it yourself.
commonchat · repo
✓ CI · secret-scanning passed
✓ CI · security-review passed
✓ CI · widget-size-budget 9 KB / 12 KB
LICENSE · Apache-2.0
SECURITY.md · responsible disclosure
Open source: license, security policy, green CI
How to sign up
Two ways in. One of them is free forever.
Hosted and self-hosted share the same core, so you can move in either direction without rebuilding your bots.
Apache 2.0
Self-Hosted
$0 / forever
The full product, no caps on bots or content. Not a crippled demo.
- Run it with Docker on your own server
- Same core code as our hosted edition
- Your data in standard Postgres
- Backup + restore procedures in the repo
Drop it on Railway, Fly, or the dusty box under your desk.
Early access
Hosted
Flat fee + AI at cost
We run the platform. AI stays at provider cost on your own key.
- chat.commonpeople.co, private beta
- Add your key, upload content, paste snippet
- Flat subscription covers the platform
- Zero markup on tokens, ever
File search uses OpenAI embeddings today, so an OpenAI key is required for that piece. An alternative provider is on the roadmap.
And lots more coming
Tell us what you want. We'll build it. Probably out of spite.
This product exists because other tools raised prices while their repos gathered dust. The roadmap works the same way: you tell us what the incumbents won't build, and we build it. On deck: custom domains, white-label branding, SSO, audit logs, outbound lead webhooks, and a second embeddings provider.
Start FreeGitHub coming soon
Or just tell us in the chat widget. Yes, the one on this page. Yes, we read the transcripts. That's literally a feature.