Krisp
Built for live chat agents and AI bots to work as one team
Krisp combines a fast shared inbox for human support agents, grounded AI bots that answer from your store’s knowledge, visual automation flows, and a per-reply cost ledger—so your team resolves conversations faster without losing control of quality or spend.
- Fast agent workspaceKeyboard-first inbox, split-send state triage, internal notes, saved replies, and search that took about 20 ms across a million messages in our tests.
- Grounded AI botsTrained on your help articles, website pages, documents, and resolved team answers.
- Visual automations10 trigger types, dry-run testing, 11 built-in plugins, and signed order lookup webhooks.
- Costs you controlEvery AI call logged to the cent with hourly and monthly hard caps, plus a free built-in engine.
Bots handle the routine. Agents handle the moments that matter.
Every incoming message—from your website widget, email, WhatsApp, Instagram, Telegram, or Facebook Messenger—moves through one predictable pipeline.
-
1. Plugins & signed order lookup
As soon as a visitor writes, the plugin chains you set up from 11 built-ins - Sentiment Radar, Intent Tagger, Lead Scorer and more - read the message and log what they find. If order lookup is on and an order ID is spotted, Krisp verifies the customer’s email with your server over an HMAC-SHA256 signed request and delivers their order details right in the chat - with the AI on or off, but never in a conversation your team is handling.
-
2. Visual automations & AI autopilot
Matching automation rules run first; otherwise the AI bot replies using your published help articles, crawled website pages, and trained Q&A. Every AI reply records its confidence score, names the source it used, and logs its exact token cost. With the AI switched off, the message goes straight to your team instead.
-
3. Clean hand-off to human agents
When confidence is below your threshold, a customer asks for a person, or an automation escalates, the conversation moves to your team with a chime and Web Push alert. Once a human takes over, all bots stay silent until you hand the thread back - and once a teammate has replied in it, until it is resolved.
A calm, high-speed desk for support teams
Support agents spend their day in the inbox, so every interaction is built for speed. Triage conversations between Unresolved, Pending, and Resolved from the thread header, toolbar dropdown, list hover actions, composer split-send button, or single-key shortcuts (e resolves, p sets pending, u reopens, r replies).
- Search (about 20 ms across a million messages, in our tests) that opens the conversation at the exact matching line, and Ctrl+K to search the whole workspace
- One-click AI reply drafts, tone rewrites,
/saved replies, internal team notes, and due-date reminders - Live visitor context: local time, location, IP address and network, current page, device, tags, past threads, and per-conversation AI spend
An AI bot you teach—never a black box
Most support bots guess when they do not know an answer or hide how they reached it. In Krisp, you control every source the AI reads, the exact confidence threshold where it hands over to a teammate, and the model that powers it.
- Feed it help articles, website URLs, documents, written answers, and resolved team conversations
- Use Krisp’s model, the free built-in engine, or bring your own key for OpenAI, Anthropic, Gemini, Mistral, Groq, DeepSeek, or OpenRouter - or switch the AI off
- Test your automation flows in dry-run mode on sample messages before switching them on
Why we built Krisp: zero surprise AI bills
We built Krisp because legacy chat tools turned every feature into an add-on tier and every AI reply into an unpredictable invoice at the end of the month. In Krisp, every AI completion is metered call-by-call with its model, input/output tokens, latency, and cost.
- Set monthly and hourly spending caps with configurable alert thresholds
- Enable the hard stop so conversations route to your human agents the moment a cap is reached
- Run the built-in local engine on any plan for $0.00 per reply
The questions we built Krisp to answer
What we chose to build, and the question to put to any other helpdesk you are weighing it against.
| Capability | Krisp | Worth asking any other tool |
|---|---|---|
| AI bot pricing & budget control | Exact token cost logged per reply, hard monthly & hourly caps, plus a $0 built-in engine | What does one AI answer cost, and what stops the spending in a busy month? |
| Bot-to-agent hand-off | Automatic silence once a thread is Pending or assigned to a person, and from a teammate’s reply until the thread is resolved | Does the bot stop on its own once a person has taken over? |
| Order & credential lookup | Live HMAC-SHA256 signed webhook to your server; orders are never copied into Krisp | Does looking up an order mean copying your orders into their system? |
| AI model choice (BYOK) | Use Krisp’s model, built-in local engine, or your own OpenAI, Anthropic, Gemini, Groq, or OpenRouter key | Can you bring your own AI provider’s key, and see the tokens each reply used? |
| Omnichannel inbox | Website widget, hosted chat page, email, WhatsApp, Instagram, Telegram bot & Facebook Messenger, with replies from Discord threads | Which channels are in the plan, and which cost extra? |
Fast, self-contained, and honest about what is built
Under the hood, Krisp is built to be reliable, auditable, and straightforward to run.
Zero-bloat architecture
Krisp runs as a single, self-contained Node.js service with zero third-party runtime dependencies on SQLite, streaming continuous backups off-server in the EU. Fewer moving parts mean faster responses and fewer supply-chain risks.
Sealed secrets & scoped data
AI provider keys, bot tokens, and other channel credentials are tested before saving and encrypted at rest with AES-256-GCM. The chat can only be framed on the websites you list, enforced by Content-Security-Policy headers.
Honest product status
Every page on this site describes what runs in production today, with no invented customer counts or quotes. Where a feature has a limit - WhatsApp’s 24-hour reply window, say - the page that names it says so.
Explore more
Give your agents and AI bots one shared desk
Start free with up to 3 teammates, install the widget in a minute, and see every conversation and AI cost in one place.