Gunjo · Business Intelligence for the AI Era
← Sticker Wall AGENT · DETAIL

Building private voice agents for SMEs using Cartesia edge and local deployment, generating 28,000 RMB monthly revenue

Workflow: Each morning, I collect leads for SMEs with outbound/inbound needs from local service communities and industry groups, o

AGENT

Key Fields

FIELD STAMPS
IndustryMarketing / Advertising
RegionChina
ScaleSME
ChannelOnline

🔧 Workflow

Each morning, I collect leads for SMEs with outbound/inbound needs from local service communities and industry groups, organizing them into a prospect list for cold calling to confirm use cases. In the afternoon, I use the Cartesia local deployment version combined with open-source LLMs to build a demo agent with a private knowledge base, tuning the voice and interruption response against the client's actual business scripts, and providing a three-day trial number with sample recordings. In the evening, I follow up with trial clients to review call data. Once signed, I charge a one-time deployment fee and a monthly maintenance fee, handling alerts and model upgrades remotely.

🛠 Setup Requirements

Requires intermediate development skills: proficiency in calling Cartesia real-time voice APIs, basic Linux server deployment, SIP trunking, and building RAG (Retrieval-Augmented Generation) knowledge bases. The tech stack includes Cartesia local/edge versions, self-hosted open-source LLMs, domestic cloud servers, and the client's existing infrastructure. The first deployment takes 2-4 weeks to troubleshoot, after which the same template is reused, allowing for delivery in 2 days per client. No self-developed models are needed; the core value lies in deployment engineering and understanding industry-specific scripts.

🧰 Toolchain

  • 🔧 Cartesia Sonic and Ink local deployment versions
  • 🔧 Cartesia Managed Agents
  • 🔧 Open-source Large Language Models
  • 🔧 Cloud servers and SIP phone lines

💰 Revenue

① SME private deployment project fees (Cash flow): Institutions like banks, medical aesthetics, and government hotlines pay a one-time fee per project, ranging from 8,000 to 20,000 RMB per client. Specific contract numbers are not publicly recorded, and the share of total revenue is unclear (based on self-reported data as of 2026). ② Recurring monthly maintenance subscriptions (Compound growth, primary revenue): Tiered pricing based on concurrent lines, 800-3,000 RMB/client/month × 15-20 clients = 12,000-60,000 RMB/month. This profile estimates 20,000-35,000 RMB/month (using 28,000 RMB as the figure), representing the stable core income (approx. 100%; based on self-reported data, not independently audited). ③ Remote O&M and line integration value-adds: Bundled with deployment; specific pricing and client coverage are undisclosed, and its impact on total revenue is not quantifiable. ④ Opportunity - Regional compliant voice agent integration: With Cartesia serving over 50,000 clients (officially announced), there is potential for individual developers to act as the local implementation layer, though this revenue is currently uncalculated.

💸 Cost

Cartesia usage-based fees range from 2,000 to 5,000 RMB/month depending on call volume. Cloud servers and phone lines cost approximately 1,500 RMB/month. Including remote O&M miscellaneous expenses, total monthly costs are kept under 6,500 RMB, maintaining a gross margin of over 70%.

⏱ Time Investment

During the acquisition phase, 4-5 hours daily are spent on cold calling, demos, and deployment. During the maintenance phase, 1-2 hours daily are spent handling alerts, upgrading models, and tuning scripts, with bulk upgrades performed on weekends.

🚀 Getting Started

Step 1: Register for a Cartesia developer account, build a demo agent that can answer booking calls using their real-time TTS interface, and verify latency. Step 2: Find a local dental clinic or beauty salon to install it for free for one week, using real call recordings as testimonials and case studies. Step 3: Use the recordings and latency data to replicate the service for similar local businesses, charging a deployment fee first, then transitioning to a monthly maintenance subscription.

🔑 Keys to Success

  • ✅ Focus on compliance (data stays on-premise) to bypass price wars with major cloud providers; short client decision-making cycles.
  • ✅ Reuse deployment templates across multiple clients; industry-specific scripts and voices accumulate, reducing maintenance costs per client over time.
  • ✅ Use 82-90ms latency benchmarks as a sales pitch; high conversion rates through live side-by-side comparisons.
  • ✅ Human-in-the-loop: Weekly manual review of call recordings to correct hallucinations and errors, ensuring safety in medical and financial scenarios.
  • ✅ Establish deep ties with client infrastructure and phone lines; high switching costs create a moat for renewals.

⚠️ 风险

  • ⚠️ Official Managed Agents product lines may eventually commoditize and squeeze individual deployment businesses.
  • ⚠️ Slow payment cycles for private clients; loss of maintenance fees if contracts aren't renewed, requiring annual contracts for stability.
  • ⚠️ Voice cloning involves authorization compliance risks; clients must sign voice authorization forms and retain proof.

📌 Real Cases

  • 📌 Cartesia officially disclosed that its Sonic model is used by companies like ServiceNow, Cresta, and Decagon to drive millions of conversations monthly, validating enterprise willingness to pay for real-time voice agents.
  • 📌 As of March 2026, Cartesia serves over 50,000 clients across customer service and companion scenarios, demonstrating that the market demand for low-latency voice agents has been validated at scale.
  • 📌 In June 2026, Cartesia released Sonic-3.5 and Ink-2, reducing TTFB to 82ms with native turn-taking detection. Alibaba Cloud entrepreneur platform records show that early versions of Sonic were already the fastest in their class at 135ms, with continuous technical iteration lowering the barrier for private deployment.