Reads your docs. Writes in your voice. Stays inside the lines you draw. $0.49 a resolution after 50 free. Half of Fin. One kill switch.

Get an AI summary

The loop

One loop. Five moves.

A customer support AI agent is only as good as the loop behind it. Kal captures what your team already knows, stays inside the rules you set, answers the queue, proves every word, and gets sharper on what you approve. Then it starts over.

  1. 01Capture
  2. 02Govern
  3. 03Answer
  4. 04Prove
  5. 05Improve

01 · Capture

Your best answers already live in Slack.

Your team solves hard problems in Slack every day, and the answer vanishes into the scroll. Kal captures those answers as they happen and turns each into a draft. Slack today, more sources soon. Nothing enters Kal until a human approves it.

Knowledge Capture
In Slack Kal is listening to #support-eng
PR
Priya #support-eng · 10:24

For the CSV export timing out past 10k rows: have them filter by date range first, then export in smaller batches. We're raising the row cap next release.

In Drafts Nothing enters Kal until you approve
Drafted by Kal Captured from Slack · #support-eng · Priya
CSV export times out on large datasets

Filter by date range, then export in smaller batches. A higher row cap ships in the next release.

In Kal The repository of sources Kal reads from

Approve the draft and it lands here, with a link back to Slack. Reject it and it never enters Kal.

Drafts first

Nothing enters Kal on its own.

Every capture lands in Drafts. A person approves it before it reaches the Knowledge Store, the repository of sources Kal reads from. Reject it and it never counts.

Provenance

Every entry keeps its receipts.

The channel, the person who posted it, and a link straight back to the original Slack message. You always know where an answer came from.

Your channels

You pick what Kal hears.

A settings page chooses which Slack channels Kal listens to. Turn on #support-eng, leave the watercooler alone. Change it whenever you like.

02 · Govern

Your AI.
Your rules.

Most AI ships as a black box. A workflow builder here. A threshold there. A bill you can't predict. Kal flips the script. Pick the character. Set the rules. Draw the lines. One screen. This one works. Try it.

Character Per topic
+ Custom

"The 429s trace to the new burst window in v3. Raise the limit or batch the calls."

Rules 3 active
Auto-resolve refunds under $50
Route angry customers to humans
Open Linear ticket on bug reports
Promise refund timelines
Guardrails Always on
Never quote prices Never mention competitors Never promise SLAs
Confidence threshold 0.82
Below: hand off Above: auto-send

What you control

Six dials. No black box.

AI you can configure is AI you can trust. Six dials. Most teams touch three. All six there when you need them.

Voice

Sounds like you. Reads like you.

Kal writes the way you write. Short or warm. Blunt or punny. Feed it three past replies and it learns the shape. Change it anytime. No retraining cycle. No "AI brand guidelines" call.

Scope

You draw the line.

Point Kal at the articles, macros, or ticket types it should handle. Block the ones it shouldn't. Refunds to humans? Done. Pricing to auto-reply? Done. Your line. Kal respects it.

Cost

Half-price by the answer.

$0.49 per resolution. Exactly half of Fin's $0.99. You only pay when Kal actually closes a ticket, nothing for handoffs. Same job, half the bill, every time.

Handoff

Smooth handoffs.

When Kal hits the edge, it hands off the full thread, the articles it consulted, and a draft of what it was going to say. Your agent picks up mid-sentence. Nothing restarts.

Knowledge

Your KB. Your truth.

Kal reads the knowledge base you wrote. Every answer cites the article it pulled from. When your docs change, Kal changes. No drift. No mystery answers. No "where did that come from."

Data

Yours. Not the model's.

Your conversations never train a public model. GDPR compliant from day one. Export everything in one click. We host Kal. You own what it reads and writes.

The keys

You hold the keys.

Kal is on. Working the queue.

Citations on every answer. Override on every reply. And one switch that turns Kal off at 2am. No ticket. No call. Go ahead, try it.

03 · Answer

Four jobs. Done before coffee.

Kal isn't a chatbot. It's the teammate who shows up early, reads everything, and leaves a draft on every ticket.

Inbox · Kal
JM
Jamie M. · 2m ago

Hey, my refund hasn't shown up yet. It's been 6 days. Order #4821.

Drafted by Kal · 0.3s Cites: Refund timing →

Hi Jamie, refunds settle in 5-10 business days depending on your bank. Order #4821 was issued on the 22nd, so it should land by Friday. I'll flag it if it slips.

Beyond the drafts

Where chatbots stop. Kal keeps going.

A chatbot waits for the next message. An agent works the queue. Kal files the bug, takes the customer's side, reads the room, and picks the right voice for the moment. The quiet work most tools skip is the work that earns the seat.

01

Integrations

Files the bug while you sleep.

Kal spots a bug in a thread and opens a Linear issue with repro steps, the customer email, and a link back. No "I'll file that for you" promises your team forgets. The work moves the moment the conversation ends.

02

Perspective

On the customer's side.

No copy-paste policy recitations. No "as per our terms." Kal argues for the customer inside the guardrails you set, flags edge cases to a human, and never gaslights a paying user.

03

EQ

Reads the room.

Kal reads sentiment, not just intent. Angry customers don't get cheerful bullet points. Confused ones don't get a wall of links. Sad ones don't get an upsell. Tone shifts to match the human in front of it.

04

Characters

Pick a character. Change it anytime.

Concierge for onboarding. Dry technical for API tickets. Warm operator for billing. Kal ships with a roster, you can write your own, and you can swap them per topic. One product. Many voices. None of them canned.

Kal takes tickets. Not jobs.

The work nobody missed goes to Kal. The work that needs a person goes to your people. That was always the idea.

04 · Prove

No black box. No fake model.

Plenty of AI support tools claim a secret in-house model. We won't insult you. Kal runs on frontier models, the best ones available, wrapped in a pipeline we built and tuned for support. Here is exactly what happens between a customer's message and Kal's reply. Every citation traces back to its source: a document, a macro, or the original Slack message it was captured from.

Customer message 01Parse02Recall03Focus04Reason05Verify06Learn Reply, action, or handoff
  1. 01

    Parse

    Reads the message. Pulls out the real ask, the intent, the entities, the mood, and strips the noise around them.

  2. 02

    Recall

    Pulls the matching facts from your Knowledge Store: the right article, the right macro, this customer's own history.

  3. 03

    Focus

    Narrows to what this ticket needs. Drops the irrelevant, keeps the few sources that actually answer it.

  4. 04

    Reason

    Weighs the answer against your policy and scope. Decides one of three things: reply, take the action, or hand off.

  5. 05

    Verify

    Checks the draft against the source before anything sends. No claim without a citation. No action without authorization. Every citation points to its source, down to the original Slack message it came from.

  6. 06

    Learn

    Every edit, override, and correction becomes signal. Kal moves closer to your voice and your rules with each one.

  7. Six stages. One message. Every reply runs the full stack, top to bottom.

The pipeline is ours. The models are the best in the world, and we swap them the day a better one ships. You get frontier intelligence without a frontier bill, and without a marketing fairytale about a model we never trained.

Kal vs Fin

Same job. Different bill.

Both answer tickets with AI. What you pay, what you control, what happens to your data. None of it is the same.

  Kal Fin
Per-answer fee $0.49 each (half of Fin) $0.99 each
Pay only on close Yes. Nothing for handoffs No. Billed regardless
Voice setup Three samples Paid onboarding
Citations Every reply Optional
Scope control Per topic, per segment Workflow builder
Handoff context Full thread plus draft Basic
Trains public models Never Subject to change

Want the long argument? Read Intercom vs SupportWire.

05 · Improve

Sharper with every answer you approve.

Kal never ships and forgets. Every answer your team approves feeds the loop. It analyzes, trains, tests, and deploys, one account at a time. Sharper because you taught it, never on its own.

01 Analyze

Reads every conversation.

Kal goes back over your account. What stalled, what your team fixed by hand, what it could have caught the first time.

Flagged conversation
Where is my refund?
Refunds take 5 to 7 business days.
Agent rewrote this reply

What stalled. What your team fixed by hand.

02 Train

Turns corrections into signal.

Your voice. Your edge cases. The answers you approved. Every edit and approval teaches Kal how your team actually answers.

The correction
Refunds take 5 to 7 business days.
Refunded today. It lands in 2 to 3 days.
Approved. This became the pattern.
03 Test

Replays before it ships.

The change runs against past conversations first. Nothing goes live that breaks what already worked.

Replay scorecard
Replayed against 214 past refund conversations
209 match the approved answer
0 regressions
5 flagged for review
04 Deploy

Pushes the win live.

Quietly, on your account. The next reply is already sharper than the last. Then the loop starts over.

Deployment
Live on your account.

The next refund reply ships the new answer.

Four steps. No finish line. The same loop runs on every account, around the clock.

How it works

Three steps. One hour.

Import to first answer in about an hour. No onboarding call. No solutions engineer assigned to your account.

01

Plug in your docs.

One click. Connect Intercom, Notion, Help Scout, and the Slack channels Kal should learn from. It indexes everything in under an hour.

02

Set voice and scope.

Paste three replies for tone. Pick the topics Kal owns and the ones it routes. Drag a confidence dial. No code.

03

Kal answers. You watch.

Every reply is logged with citations, confidence, and edit history. Override any answer. Correct once, retrain forever. Your team stays in the loop.

Why we built Kal

AI should earn its keep.
Not pad the bill.

When Intercom launched Fin at $0.99 per resolution, founders ran the math and panicked. 10,000 tickets a month meant a $10,000 AI line item. An AI agent should scale your team. Not your problems.

Kal is the Intercom Fin alternative at half the per-resolution price: $0.49 each, where Fin is $0.99. Same job, half the bill. You pay only when Kal closes the ticket; nothing for handoffs. Scope what it answers with toggles, not a consulting engagement.

We built Kal for the team that explains the bill to the CFO. For the engineer who wants to read the citation. For the support lead who wants to kill the switch at 2am without a support ticket of their own.

A public complaint about Intercom's per-resolution Fin pricing
The kind of bill we don't write.

Designed with intention

Three moods. One orb.

Kal is a quiet ring of light. It brightens when you speak, and runs the full spectrum when it works. Go on, try it.

A quiet pulse of light.

FAQ

Questions. Honest answers.

Written by the team that built it. Sales fluff not included.

Two things matter. Pricing: Fin charges $0.99 per resolution. Kal is $0.49, exactly half, and only when Kal closes the ticket. Control: Fin is a workflow builder with AI bolted on. Kal is AI with guardrails. Scoped by topic. Configurable to your voice with three sample replies. Transparent on every answer with citations. Full breakdown: Intercom vs SupportWire.

Frontier models, the best general-purpose models available, wrapped in a retrieval-and-verification pipeline we built for support: Parse, Recall, Focus, Reason, Verify, Learn. We don't claim a secret in-house model, because honesty is the differentiator. When a better model ships, Kal uses it, and you don't pay more.

Yes. Our customer support MCP server lets Claude, ChatGPT, Cursor, and VS Code list conversations, reply in-thread, and update customers over OAuth, with no API keys to copy.

Drafts replies from your KB. Summarizes long threads. Closes the repeatable end to end. Routes the rest with full context. Not a chatbot reciting canned answers. It reads the ticket, reads the docs, writes a reply you can send, edit, or override.

Kal watches the Slack channels you choose and captures the answers your team gives. Each capture becomes a draft. A person approves it before it enters the Knowledge Store, the repository of sources Kal reads from. Slack today, more sources soon. It is not a scraped knowledge base you can't see into: every entry links back to the exact Slack message it came from.

Completely. You pick which Slack channels Kal listens to on a settings page. Everything it captures lands in Drafts, and nothing reaches a customer until you approve it. Reject a draft and it never enters Kal. You teach it on purpose, one approval at a time.

Minutes. Plug in your knowledge base. Paste three past replies. Kal starts drafting. Most teams have Kal handling 30% of their inbox by end of day one. The more you correct it, the better it gets. Every edit is training data.

Kal reads your data to answer your customers. Your conversations never train a public model. Ever. We're GDPR compliant from day one. Export everything in one click. If you cancel, we delete your data within 30 days. You can read the full privacy policy.

Kal is included at $19 per seat per month, with unlimited free lite seats. The first 50 AI resolutions each month are included, then $0.49 each, half of Fin's $0.99. You only pay when Kal actually closes a ticket, nothing when it hands off to a human. See the full pricing page.

Yes. Per-topic, per-segment, or all at once. Kal is a teammate your team manages, not a black box you subscribe to. Want Kal to handle only onboarding questions and nothing else? One toggle. Want to pause during a sensitive incident? One toggle. You're always in charge.

Updated June 2026

Try Kal. See the citations.

14-day trial. No card. Kal is answering tickets before your coffee cools.