Ninjō AI lets a marketing agency run one AI sales agent per client across Instagram, WhatsApp and six more channels, with per-client autonomy rules and an MCP server that drives the whole operation from Claude or ChatGPT.
Used this tool? Rate it
Used this tool? Rate it
Ninjō AI is an AI agent platform that runs direct-message sales conversations for marketing agencies across a whole portfolio of client accounts. It handles every DM across that portfolio — qualifying leads, pitching offers and closing sales in each client's voice — and positions that as a recurring revenue stream the agency resells. The product is operated by Katalyst Arc, LLC, the entity named in the privacy policy.
The unit of work is the agent, not the inbox. An agency builds one agent per client, trained on that client's voice, offers and audience, then supervises all of them from a single agency dashboard designed to cover anywhere from five clients to five hundred. That framing makes it portfolio infrastructure rather than a single-brand chatbot.
Two control surfaces exist in parallel. One is the web dashboard. The other is a Model Context Protocol server that lets an operator create agents, tune prompts, review pipelines and pull reports by chatting with Claude, ChatGPT, Claude Code or Codex, without opening the dashboard at all. That second surface is the product's most distinctive claim.
Each agent is trained on an individual client's voice, offers and audience, so replies follow that client's communication style rather than a shared template. The agent runs the full sales loop — qualifying leads, handling objections and booking calls. Ninjō also analyzes every conversation to spot patterns, improve objection handling and increase conversion rate over time, which positions the agent as tuned rather than static.
Agents run on Instagram, WhatsApp, SMS, Telegram, Facebook, TikTok, Skool and Slack, with one unified inbox per client and one agency dashboard across all of them. The two vendor pages do not agree, however: the features page names all eight channels, while the self-serve plan card on the pricing page lists only Instagram, WhatsApp, Facebook, TikTok, Skool and Slack. Telegram and SMS appear on the capability page but not in that plan's bullets, so confirm channel availability per plan before committing.
Autonomy is configured per client, not globally. The three documented settings are fully autonomous, partially supervised and human-first, and every client account gets its own rules. Escalation triggers are likewise per-account — high-ticket leads, stalled conversations, or any trigger the operator defines. When a trigger fires, the operator receives a real-time alert on WhatsApp or Slack with the lead's name, context and a one-click action, and can take the conversation over.
The dashboard reports agent performance, content insights and audience patterns across the portfolio, and exposes real-time conversations with filters for qualifying leads. Funnel intelligence shows where leads drop off, which agents are active at each stage and which objections stall each pipeline. Alerting adds daily growth digests, hot-lead alerts, objection patterns and funnel health per managed client. The metrics named for per-client review are conversion rates, response times and best offers.
Outbound runs on approved templates over SMS and WhatsApp, paired with nurturing sequences that reactivate cold leads, and the agent takes every reply rather than handing the thread back to a form. Through the MCP surface the same outbound templates can be sent in batches. The approved-template requirement is not a Ninjō invention: messaging platforms constrain business-initiated messages outside an active conversation window, so template approval is the normal path to a lead who has gone quiet.
The MCP server works natively with Claude, ChatGPT, Codex and Claude Code, plus any other MCP client. Five action categories are named: managing conversations, pulling pipeline and audience reports, handling handoffs, sending outbound templates in batches, and generating per-client dashboards shareable with the client's own team.
The third-party launch listing adds capabilities the marketing pages do not foreground: versioned changes with instant rollback, synthetic-conversation testing, follow-ups, keyword triggers and a built-in CRM. Versioning and synthetic testing matter more for an agency than for a single brand, because a prompt change that degrades replies ships into several clients' revenue at once.
Five jobs are named across the product pages: closing sales, customer support, automatic reports, outbound SMS and WhatsApp, and nurturing. The suite is described as designed to work individually or together, so one client might run only inbound qualification while another adds reactivation.
The reporting job is the most easily underrated. Because a per-client dashboard can be generated and shared with the client's own team, the reporting output doubles as the deliverable an agency shows at a monthly review — the agency is not only automating DMs but producing the evidence that the automation worked.
The documented MCP path is four steps. First, add Ninjō's MCP server to Claude or ChatGPT, described as one click with no code and no config files. Second, describe what you want in plain language — launch a campaign, check the funnel, reply to a lead. Third, Ninjō executes it: the AI calls the matching Ninjō tools, runs the action and confirms what happened in real time. Fourth, iterate in the same conversation.
The resulting loop is summarized as create, test, analyze and improve. The illustrated example runs the whole sequence in one exchange: an operator asks for an agent that qualifies leads and sends a payment link, the assistant creates it, connects it to Instagram DMs, verifies that it replies, and then wires a Slack notification on purchase.
The self-serve plan advertises a five-minute setup and describes building and running agents yourself, from either Claude or the dashboard. The sequence is to connect the client's messaging account, describe that client's offer, and adjust tone and responses before switching the agent live.
The white-label route is not self-serve at all: that tier includes partnership setup and guided onboarding with a dedicated success manager, implying a scheduled implementation rather than a same-day launch.
The product's framing is narrow and consistent: it is built for marketing agencies scaling their clients' sales and operations with AI. Every major surface — per-client agents, per-client autonomy rules, per-client dashboards, one agency dashboard over all of them — assumes a portfolio rather than a single brand.
It is a weaker fit for a single in-house brand team, which pays for portfolio machinery it will not use; for teams whose pipeline lives in email or outbound calling rather than social DMs; and for buyers who need a published contract before purchase, since no public terms of service document was reachable this round.
Ninjō is a hosted web platform. The documented control surfaces are the web dashboard and the MCP server, and the public site does not advertise a native iOS or Android app for operators — mobile appears as the destination for alerts, delivered into WhatsApp and Slack. The white-label tier's bullets do mention handoff and alerts arriving on WhatsApp, Slack and a Ninjō App.
Client-side reach is defined by messaging channels rather than operating systems: Instagram, WhatsApp, SMS, Telegram, Facebook, TikTok, Skool and Slack. Skool's presence is a signal about who this is for, since it is a community platform used by course and coaching businesses rather than a general consumer channel.
The named integrations are ManyChat, n8n, Tienda Nube, GoHighLevel, Calendly, Slack and Hotmart. Calendly covers booking, Tienda Nube and Hotmart cover Latin American commerce and course selling, and n8n covers custom automation — together describing the business model the product grew out of. No CRM connector is named beyond the built-in CRM reported on the launch listing, so if your agency reports into Salesforce or HubSpot, treat that as unverified.
| Plan | Headline price | Usage cost | Trial and free allowance | Support model |
|---|---|---|---|---|
| Self-serve | $150 / month | $0.035 per message sent | 14-day free trial; 500 free messages, no credit card required | Self-directed, from Claude or the dashboard |
| White Label Agencies | Custom | Not published | Not published | Partnership setup, guided onboarding, dedicated success manager |
The self-serve plan bundles unlimited agents operated from a chat, follow-ups, triggers and automatic alerts, with create, analyze, improve and automate all included. What is not bundled is volume: the per-message fee sits on top of the subscription, so cost scales with conversation volume rather than with client count. The plan that allows unlimited agents still meters every message those agents send.
The free-message allowance is the one figure where sources disagree. The pricing page states 500 free messages with no credit card required, while the third-party launch listing advertises 1,000 free messages with a launch code — the standing offer versus a promotional one.
The white-label tier publishes no price at all. It adds partnership setup and guided onboarding, management of all clients from one account, a dedicated success manager, per-client autonomy rules and levels, handoff and alerts across WhatsApp, Slack and the Ninjō App, and custom integrations and flows. Entry is by booking a call, so terms are negotiated rather than listed.
For context on what that negotiation is anchored against, independent directories put published white-label tiers in a comparable band: one agency roundup lists competing white-label plans at $297 and $497 per month, and a separate messaging-platform roundup lists entry prices from about $15 to $399 per month across ten lead-generation tools. Neither list includes Ninjō, so these are category reference points, not like-for-like comparisons.
The clearest documented boundary concerns integrations. The privacy policy states that it does not apply to third-party websites, applications or integrations accessed through the Services, and that those third parties are responsible for their own privacy practices. Since the entire product runs on other companies' messaging platforms, the guarantees you are reading cover Ninjō's own handling and stop at the channel boundary.
Security is stated with an explicit limit rather than a promise: the policy says no system is completely secure, and commits to notifying affected users as required by law if a breach materially affects them. That is a notification commitment, not an assurance.
Data-subject requests do not route to Ninjō by default. Where personal data is processed as Customer Data, the policy directs the request to the customer who controls that data, because Katalyst acts as processor in that context. For an agency this is the operationally important sentence: end-user requests from your clients' audiences land on you, not on the vendor. The published response window for requests Katalyst does handle is typically thirty to forty-five days.
Three further boundaries are worth planning around. Personal data may be stored and processed in any country where Katalyst or its service providers operate, including the United States, with Standard Contractual Clauses named as a safeguard. The Services are intended for business and professional use and are not directed to anyone under eighteen. And the US opt-out of advertising-related sharing is not a self-serve toggle — the policy directs you to record that election in the Statement of Work, a contract action rather than a settings change.
Two evidence gaps should also weigh on a purchase decision. No public terms of service, refund or cancellation document was reachable this round; the footer links only pricing and the privacy policy. And the product is young — the launch listing is dated August 2026 — so the vendor's headline figures have no independent counterpart yet, and they disagree between sources: the homepage cites more than 120 live agents while the launch listing cites more than 150 in production. An automated domain-reputation scan returns a positive overall verdict while flagging hidden WHOIS ownership and low traffic rank, but such scans measure registration signals, not product behaviour.
It runs direct-message sales conversations on behalf of an agency's clients. One agent is built per client, trained on that client's voice, offers and audience, and it qualifies leads, handles objections and books calls while the agency supervises the whole portfolio from one dashboard.
The capability page lists Instagram, WhatsApp, SMS, Telegram, Facebook, TikTok, Skool and Slack, with one unified inbox per client. The self-serve plan card lists a shorter set that omits Telegram and SMS, so confirm availability for the plan you intend to buy.
The self-serve plan is $150 per month plus $0.035 for each message sent, after a 14-day free trial that includes 500 free messages with no credit card required. The white-label tier is custom-priced and requires booking a call.
No. The MCP server works with Claude, ChatGPT, Codex and Claude Code, and lets an operator create agents, review pipelines, reply to leads, handle handoffs and pull reports from the chat client. The dashboard remains available as the alternative surface.
Yes, and takeover rules are set per client. Autonomy can be fully autonomous, partially supervised or human-first, escalation triggers are defined per account, and alerts arrive in WhatsApp or Slack with the lead's context and a one-click action.
The privacy policy states that identifiable Customer Data is never used for AI model training, profiling or commercial analytics, and that only aggregated, de-identified or behavioural data may be used to improve model accuracy, safety and performance. Google Workspace user data is likewise never used to develop or train generalized or foundational models.
Katalyst Arc, LLC acts as data processor for Customer Data under the customer's documented instructions, and as controller only for billing, account administration, security logs and aggregated analytics. Requests about Customer Data are directed to the customer who controls it, which in an agency arrangement means the request reaches you rather than the vendor.
