CometChat

CometChat sells four separately-priced things that share one conversation layer, and it is worth knowing which one you are shopping for before you look at a price.

Chat & Messaging — the original product. SDKs and UI kits for React, React Native, Angular, Android, iOS, Flutter and Unreal, plus REST APIs, moderation, push/email/SMS notifications, webhooks and analytics. Priced by Monthly Active Users (MAU) and Peak Concurrent Connections (PCC).

Voice & Video Calling — usage-priced per minute, and usable without the chat product (the vendor's pricing FAQ says so explicitly).

AI Agents — CometChat-hosted agents running inside those conversations, billed by plan plus consumable "Agent Credits".

Bring Your Own Agent (BYOA) — the chat infrastructure only, with your own Agno / CrewAI / AG2 / Mastra / LangGraph agent plugged in. Priced by MAU like the chat product, not by credits.

The AI Agents and BYOA split is the most consequential distinction on the pricing page and the easiest to miss: if you already have an agent, BYOA is the line you want, and it does not consume Credits at all.

This is the finding that most distinguishes CometChat from its direct competitors, and it is unusually well-evidenced for a marketing claim because you can hit the endpoints yourself.

CometChat ships a first-party MCP server at https://mcp.cometchat.com/mcp — read-only, no account, no API key, no auth — that serves its documentation to any MCP client (Claude Code, Cursor, Windsurf, VS Code Copilot, Codex). The repository is Apache-2.0. Hitting https://mcp.cometchat.com/health on 16 August 2026 returned {"status":"ok","service":"cometchat-mcp","version":"0.1.5","indexReady":true,"bundles":10,"sessions":53882} — so it is live, and the session counter is a real (if vendor-reported) usage signal rather than a landing page.

Separately it publishes @cometchat/skills (MIT, on npm), installed with npx @cometchat/skills add, which writes framework-specific integration skills into your coding agent — 13 skills for web, 31 for Android, 28 for Flutter — and a /cometchat dispatcher that detects your stack and pulls the right family on demand.

And its docs are published as markdown twins: every page has a .md sibling, there is a llms.txt index, and — most unusually — a pricing.md written explicitly for LLMs, complete with a section of "Agent interpretation rules" telling a model how to quote the tiers and, notably, "If a public price cell says Get in touch or Contact us, do not invent a price."

Whether this makes the product better is a separate question from whether it makes the integration faster. It clearly makes the integration faster, and the pricing.md in particular is a level of pricing disclosure most of this category does not offer. Caveat on scale: @cometchat/skills drew 385 npm downloads in the month to 9 August 2026 — the capability is real, the adoption of it is early.

The vendor's headline numbers — "113,000+ apps", "126M+ users", "7M+ messages sent every day" — are unverifiable claims. These are not:

npm downloads, month to 9 August 2026: @cometchat/chat-sdk-javascript 124,652; @cometchat/chat-uikit-react 41,785; @cometchat/chat-uikit-react-native 32,064.

GitHub stars, 16 August 2026: react-native UI kit 867, React UI kit 855, Android 699, iOS 572, Flutter 518. The React UI kit has 400 forks and 2 open issues.

Maintenance: every current SDK and UI kit repo in the org had been pushed to within the last month when checked on 16 August 2026. This is not an abandoned codebase. (The older cometchat-sample-app-* demo repos are stale — Angular last pushed February 2025, Vue January 2025 — but those are samples, not the product.)

That is a real, working, mid-size developer footprint. It is an order of magnitude below the largest players in the category, and it is concentrated in the JavaScript SDK rather than the UI kits, which suggests a good share of users take the SDK and build their own interface.

All figures below are from CometChat's own pricing page and its machine-readable pricing.md, checked 16 August 2026, and cross-checked against the rendered page payload.

Chat, annual billing (shown as monthly equivalent):

Monthly billing is roughly 25% higher (10k MAU: Basic $373.75, Advanced $498.75, Enterprise $1,248.75).

The comparison that matters. Stream's published pricing, checked the same day, puts Start at $399/mo annual for 10,000 MAU, and moves multi-tenancy, advanced search, HIPAA and message translation up into Elevate at $599/mo annual. CometChat's Advanced includes all four at $399/mo annual for the same 10,000 MAU. On that specific comparison — and it is the like-for-like one — CometChat is roughly a third cheaper. Overage rates are effectively identical ($0.10/MAU and $1/concurrent vs Stream's $0.09 and $0.99).

CometChat also lists messages, message storage and file storage as unlimited on every plan, where Stream caps stored messages at 5M (Start) / 10M (Elevate) with overages.

Where it goes the other way. CometChat's free Build plan is 100 users and 25 concurrent connections; Stream's free tier is 1,000 MAU and 100 concurrent. If you are prototyping or running a small community, that is a ten-fold difference and it pushes you to a paid plan much sooner.

Two things to size before you commit. Concurrency is included at 5% of MAU on Basic and Advanced (10% on Enterprise) and overage is billed at $1 per concurrent connection — if your app is bursty, or is a live-event or classroom product where everyone connects at once, the PCC line can dominate the MAU line. If you blow through both, CometChat bills the higher of the two, not both. And message retention is 6 months on every plan, including Enterprise; longer retention is an extra cost that is not published. Storage is unlimited, API access to it is not.

Voice and video are pay-as-you-go and cheap on paper: $0.001/min voice, $0.003/min video, $0.006/min recording, with 10,000 free minutes on the Build plan.

Published prices (16 August 2026): Web-only is $0/mo with 1,000 starter credits and pay-as-you-go bundles from $25; Core is $99/mo annual ($123.75 monthly) with 2,500 monthly credits; Plus is $999/mo annual ($1,248.75 monthly) with 25,000 monthly credits. Top-up bundles are $25 for 2,500 credits, i.e. $0.01 per credit.

What is not published anywhere — not on the pricing page, not in pricing.md, not in the docs index — is how many Credits a request actually costs. The vendor's own FAQ says: "Each model request costs a certain number of Credits. The cost per credit is determined by the LLM selected. Some replies may require multiple requests (tools/actions), so a single reply can use multiple Credits." Which LLMs are available, and what each costs in Credits, is not established from public sources.

The practical consequence: you cannot forecast the bill for a tool-using agent before you run it. Core's 2,500 included credits are worth $25 at the top-up rate, so the great majority of the $99 is platform, not inference — meaning for any agent doing real multi-tool work, top-ups are likely the dominant cost and you will discover the rate empirically. Credits also expire: monthly allowances do not roll over, top-ups expire three months after purchase, and if you run out the agent simply stops responding unless auto-recharge is on.

Other stated limits worth knowing. The SDK docs state plainly that "Agents only respond to text messages." The OpenAI-powered contextual moderation and insights features require you to bring your own OpenAI subscription — a third-party cost on top. HIPAA/BAA on the AI Agents Plus plan is listed as "coming soon", not available.

If you already have an agent, the BYOA line sidesteps all of this: Sandbox is free to 1,000 MAU, Prime is $1,999/mo annual ($2,498.75 monthly) including 50,000 MAU, with no Credit meter — you pay your own model provider directly. For a team with an existing Agno/CrewAI/LangGraph stack, that is the more predictable purchase, though $1,999/mo is a substantial step up from the $399 chat plan and there is no published tier between Sandbox and Prime.

Unlike the obvious cloud-only comparisons, CometChat documents a genuine on-premise deployment — Docker Swarm (to roughly 200k MAU), Kubernetes above that, bare metal, VMs or your own AWS/GCP/Azure account, plus an air-gapped option and a separate air-gapped internal-messenger product (CometChat AIR). The docs describe the actual architecture — TiDB, MongoDB, Redis clusters, Kafka — rather than gesturing at it.

What makes this section worth reading is that the sizing is published, which lets you cost it honestly:

Plus 500 GB–2 TB SSD, and 2–3× that again for backups. A dev instance needs 8 vCPU / 16 GB / 100 GB SSD. Stated setup time is 1–2 weeks for standard production, 3–6 weeks for HA or air-gapped.

Read that against the $399/mo cloud plan at 10k MAU. Thirty-two vCPUs and 64 GiB, run properly with backups and on-call, is unlikely to come out cheaper at that scale once staff time is counted — self-hosting here is a data-sovereignty and compliance decision, not a cost-saving one. It starts making economic sense much further up the MAU curve, where cloud pricing goes to "get in touch".

The limitation to weigh: voice and video are not available on-premise. The vendor's own page states "On-prem support is actively being developed and will be available soon." Chat, moderation, webhooks and notifications are described as at feature parity with cloud; calling is not. Air-gapped deployments also lose external push notifications. If you need calls inside your own perimeter, this is not currently the product.

On-premise pricing is not published.

Stated on the vendor's security page: SOC 2 Type 2, ISO 27001, HIPAA with BAA available, GDPR, PIPEDA. TLS/SSL in transit; AES-256 at rest in their private cloud. We did not obtain the underlying audit reports — these are vendor statements, and a buyer should ask for the SOC 2 report and the ISO certificate directly.

No end-to-end encryption is claimed on that page, and none appears in the pricing feature matrix. That is a coherent design choice for a platform that does server-side moderation, search indexing and AI summarisation — all of which require plaintext access — but if E2EE is a requirement, this is the wrong architecture, not a missing feature.

HIPAA/BAA is gated by plan: Advanced and Enterprise chat, and Prime BYOA. It is "coming soon" on AI Agents Plus. SSO (SAML/LDAP) is Enterprise-only on chat and an add-on elsewhere — worth pricing in if you are a mid-market buyer who assumed it came with Advanced.

The uptime claim. CometChat markets "99.999% uptime SLA". We could not find a published SLA document — cometchat.com/legal-sla returned an empty page when fetched, and the Terms of Service reference "service level commitments" without stating them. Treat the 99.999% figure as a sales-stage question, not a documented commitment.

What the operational record does show, from the public status page on 16 August 2026: every component reporting 100% uptime, and the incident history for June–August 2026 consisting almost entirely of scheduled maintenance windows rather than outages. The most recent service-affecting entry we found in the older history was a degraded-performance incident on 13 February 2026 lasting roughly 23 minutes. That is a good record. The flip side is that scheduled maintenance windows are frequent — a dozen-plus across June and July — so ask how those are communicated and whether they are service-affecting for your region.

Data residency is not documented publicly. The marketing copy references 35+ edge locations; where message data is actually stored, and whether EU or regional residency is available on the cloud product, is not established from public sources. Ask.

Worth stating plainly because it is easy to misread. The UI kits are on GitHub with full source, which reasonably reads as open source at a glance. They are not. LICENSE.md in the React UI kit is three lines: "Copyright (c) 2024 CometChat Inc. License agreement: https://www.cometchat.com/legal-terms-of-service", and GitHub's licence detector accordingly returns NOASSERTION.

Those Terms grant "a non-exclusive, limited, revocable right to access and use the CometChat Technology as necessary for your use of the Services", and reserve to CometChat all right, title and interest in the technology "and all improvements and derivatives of the foregoing".

In practice teams do fork and customise these kits, and the documentation encourages customisation — this is normal commercial-SDK licensing, not a trap. But if your plan involves heavily modifying the UI kit, or you have a policy that treats "it's on GitHub" as "it's permissively licensed", have counsel read the grant. Note that the MCP server (Apache-2.0) and the skills package (MIT) are genuinely open source — the licences differ per repository, so don't generalise from either direction.

There is very little arms-length user evidence on CometChat, and what exists is easy to over-read.

Trustpilot shows 4.7/5 — but from 47 reviews accumulated since the profile was claimed in April 2016. The transparency page, checked 16 August 2026, shows one review in the past twelve months (organic, 5-star, not solicited). A 4.7 built over a decade tells you little about the product as it exists today, and the near-total absence of recent reviews tells you the company does not run review campaigns — which is to its credit as a signal of authenticity, and useless to you as evidence.

G2 returns 403 to automated access, so we could not verify any G2 rating or review count and are not repeating one.

The remaining first-page search results are SEO review farms with no verifiable methodology. We deliberately cite none of them, and you should discount them too.

Named customers. CometChat publishes case studies for Pangea.ai, Randstad, Bitdefender and Snaphunt, and displays a wider logo wall including Deloitte, Kahoot, Acast, AutoNation, NoBroker and Times Internet. Four documented case studies against sixteen-plus logos is a normal ratio; treat the unbacked logos as "has some commercial relationship" rather than "runs this in production at scale".

Support is the area where absence of evidence is most awkward, because it is also where the pricing gap sits. Standard ticketed support is included on every plan. Guaranteed first-response times, a private channel and solution-engineering resource require a paid support plan — and the price of that plan is not published; the services page's own FAQ question "How much does paid support cost?" resolves to a contact-sales path. If response-time guarantees matter to you, get that quote in the same conversation as the platform quote, because it is a real line item you cannot currently estimate.

CometChat announced $6.5M in strategic funding from existing investor Run Ventures on 17 March 2026, stating total funding to date of $21.1M. No revenue, ARR or customer-count figures were disclosed in that release.

The stated use of funds is explicitly the AI pivot. Cofounder CEO Anuj Garg: "The current fund raise is specifically done to accelerate our AI play. Our GTM in this vertical would be to target high-transaction verticals like wellness & beauty, home services, e-comm, hospitality & food service." The investor's quote describes the AI platform as being in pilot — "piloting its AI platform with encouraging early signs as enterprise customers have come onboard" — which is a fair characterisation to carry into your evaluation: the chat business is a decade-old, funded, actively-shipping product; the AI Agents product on top of it is newer and, on the investor's own description, still moving pilots into production.

That maps neatly onto the buying advice. If you are buying chat infrastructure, you are buying a mature thing. If you are buying the hosted AI agent platform, you are an early customer of a product whose per-request economics are not yet published, and you should price and contract accordingly.