Kalim · 7 min read

Kalim vs Crisp vs Intercom

Every comparison page is written by one of the products. This one is too, so here is the useful part first: if you need a mobile app, run on WooCommerce, or have a security review that asks for SOC 2, stop reading and stay where you are. Kalim does not do those things yet. What follows is for everyone else.

Four of those ten rows go against us, and they are the ones to weigh first: if any of them is a requirement rather than a preference, the rest of this page will not change your mind.

The pricing model, not the price

The single largest difference is structural. Kalim charges one price for the whole workspace; most established support tools charge per seat, per resolution, or both. That difference compounds with the size of your team rather than staying constant.

What Kalim charges, exactly:

Run that against a per-seat plan with your own numbers. A tool at $25 per agent is $50 a month for two people and $250 for ten; Kalim is $19 either way. The crossover is somewhere around the first or second hire, which is exactly when a growing team is deciding whether it can afford to put another person on the queue.

Check the current prices yourself before you decide. We are not quoting anyone else's figures on this page, because support pricing changes every few months and a number we wrote in September would be wrong by spring. Two things worth pricing carefully on whatever you are comparing: whether AI answers are billed per resolution on top of the subscription, and whether the seat count includes people who only answer occasionally.

What the AI is allowed to do

Every support product ships an AI now. The difference worth asking about is what happens when it does not know.

Kalim's answer route runs a citation gate: the model must point at the help-centre article its answer came from, and an answer it cannot attribute is never sent. The conversation goes to a person with the draft attached instead. There is no setting to turn this off, which is a deliberate limitation — it means the AI cannot improvise a refund policy, and it also means a thin help centre produces more escalations rather than more confident guesses.

The practical question for any tool you are evaluating: ask its AI something your documentation does not cover, and see whether you get an admission or an invention.

Language

Most tools treat Arabic as a translation of the interface. The messages still lay out left-to-right, punctuation lands on the wrong side of the sentence, and an order number inside an Arabic sentence scrambles.

In Kalim each message carries its own direction, so an Arabic question and an English reply sit correctly in one thread, and the agent's inbox itself runs in Arabic, French or English with the layout mirrored. Arabic conversations can be routed to a model chosen for Arabic while English stays on another.

If your customers write in one language, this is worth nothing to you and you should weigh the other differences instead. The guide on answering in Arabic and English has the five-minute test we use for whether a tool was ever really tried in RTL.

Where your data and your model run

Kalim speaks the OpenAI-compatible chat API that Ollama, vLLM and TGI serve, so an open-weights model on your own hardware is a first-class engine rather than a fallback: conversations never leave your machines. You can bring your own Anthropic key instead and compare the two side by side on cost, latency and how often the citation gate withholds an answer, before committing to either.

Tenancy is enforced by Postgres row-level security rather than by application code remembering to filter, channel credentials are encrypted at rest and never returned by the API, and private notes are never delivered to a visitor's browser.

Where an older tool is the better choice

Four cases, and they are not hypothetical:

• You need a native mobile app. Kalim runs in a browser and works on a phone, but there is no iOS or Android app, so there are no push notifications when the tab is closed. For a team that answers on the move, that is disqualifying.
• Your shop is not Shopify. Order lookups — the thing that turns "where is my order" from a search into an answer — work with Shopify today. WooCommerce and the rest are on the list, which is not the same as being in the product.
• You depend on a large integrations marketplace. Kalim has six channels and one commerce integration. If your workflow routes through a long tail of third-party apps, you will feel the gap immediately.
• You have a formal security review. Kalim is young and has no SOC 2 report to hand a procurement team. The architecture is defensible and documented, but a document is what they will ask for.

If you do migrate

1. Move the help centre first. It is what the AI answers from, so until those articles exist Kalim escalates almost everything. Rank your articles by how often the question comes up and move the top ten before you switch anything.
2. Run both for a fortnight. Put Kalim on one page — the order tracking page is a good one — rather than the whole site, and keep the old widget everywhere else.
3. Expect history to stay behind. Past conversations do not import. Keep the old tool readable for a month; almost nobody needs a two-year-old thread, but somebody always needs last week's.
4. Check the numbers you actually care about — first reply time and how many conversations got answered without a person — after two weeks, not two days.