The Solo AI Agency Tech Stack: 8 Tools, Zero Full-Time Hires

Running an AI agency solo is possible in 2026 if you have the right stack. The wrong stack means manually managing datasets, chasing clients for approvals, and doing client work that could be systematized. The right stack means one person can realistically manage 6-8 active clients at retainer rates.

These are the 8 tools that make it work.

Tool 1: Ertas — Fine-Tuning and Model Management ($69.50/mo, Agency Plan)

Ertas is the operational center of a solo AI agency. Every client project lives here: their datasets, their training runs, their model versions, their evaluation results. The Agency plan gives you 10 client-labeled projects with full isolation, which covers most solo agencies.

What it handles that you would otherwise need to build yourself:

• Dataset upload and validation (catches format errors before wasted training runs)
• Training job management (queue, monitor, compare multiple runs)
• Per-client model versioning (know exactly which GGUF file is deployed for each client)
• Evaluation tooling (bulk eval, auto-evals that catch regressions before client review)
• GGUF export (one click to the deployable file)

The 3 concurrent job slots mean you can retrain multiple clients in parallel instead of queuing them. At 6-8 clients on a quarterly retraining schedule, this eliminates week-long delivery delays.

Tool 2: Ollama — Local Inference Server (Free)

Ollama runs your fine-tuned GGUF models on any VPS. It handles model loading, serving, and the OpenAI-compatible API that your clients' systems call. You typically run one Ollama instance per client (for isolation) or a shared instance with per-request model routing.

Ollama is free and runs on a $14-26/month Hetzner VPS per client. At 8 clients, that is roughly $112-208/month in VPS costs — well within the margin of even modest retainer rates.

Practical setup time: 20-30 minutes per client, including Ollama installation, GGUF loading, and API endpoint verification. Repeat for each new client deployment.

Tool 3: n8n — Workflow Automation ($10-20/mo self-hosted)

n8n is the automation backbone. A solo agency without automation is a time-capped business — everything requires your active attention. With n8n, you automate:

• Client report generation: Pull performance metrics from your monitoring system weekly, format as a report, send to client email automatically
• Training data pipeline: When client systems export new interaction data, n8n picks it up, formats as JSONL, validates, and stages it for review
• Alert routing: When model performance drops below threshold, n8n sends you a Slack message with the specific metric and affected client
• Onboarding workflows: When a new client signs, n8n creates the project folder, sends the data collection guide, and schedules the first check-in

n8n has a native Ollama node, which means your automation workflows can also use your fine-tuned models for tasks within the automation itself (enriching data, classifying inputs before routing, generating summaries of client reports).

Self-host on a $10/month VPS (separate from client inference VPS). No per-task fees, no data leaving your environment.

Tool 4: Linear — Project Management ($0-8/mo)

Linear tracks all work across all clients. For a solo agency, the key features are:

• Client as project: Each client is a Linear project. Open issues are active deliverables; closed issues are history.
• Weekly planning: Every Monday, review all open issues across clients. Prioritize based on retainer cycles and deadlines.
• Async communication records: Log all client decisions as issue comments. When a client asks "why did you retrain in February?" you have the answer in Linear.

Linear's free tier is sufficient for 3-4 clients. The paid plan ($8/user/month) is worth it at 5+ clients for better cross-client visibility.

Alternative: Notion with a database-per-client structure. Linear is faster; Notion is more flexible. Either works.

Tool 5: Stripe — Payments and Subscriptions ($0 + 2.9% + $0.30)

Stripe handles retainer billing without any additional tooling. Set up a subscription product for each retainer tier, send the client an invoice link, and billing is automatic each month. Stripe's customer portal lets clients update payment methods without involving you.

For project-based billing: Stripe invoices with milestone-based items. The client pays online; funds arrive in your bank account in 2 business days.

Time saved vs manual invoicing: 4-6 hours/month for a 6-8 client agency. At your billing rate, that is real money.

Tool 6: Loom — Async Client Communication ($12.50-15/mo)

Solo agencies cannot afford to be in meetings all day. Loom lets you communicate with clients asynchronously:

• Monthly model performance review: Record a 5-7 minute Loom walking through the metrics, what changed, and what you plan to do. Send instead of scheduling a call.
• Model handoff: Record the deployment walkthrough so the client has a reference. Eliminates support questions.
• Training result explanation: Record yourself walking through the evaluation results when delivering a new model version. More personal than a written report, less time-intensive than a call.

Most clients prefer a well-recorded Loom to a scheduled call because it fits their schedule. Your response rate to Loom updates is typically higher than email.

Tool 7: Cal.com — Scheduling ($0)

Cal.com is open-source and self-hostable. The free cloud version is sufficient for most solo agencies: clients book calls directly without the email back-and-forth. Set availability windows that protect your focus time (e.g., calls only Tuesday and Thursday, 2-4pm).

Connect to Google Calendar or Apple Calendar for automatic conflict detection.

Time saved: 30-60 minutes per client per month on scheduling coordination.

Tool 8: GitHub — Model Versioning and Script Management ($0)

Git is not optional for a solo agency managing multiple client models. GitHub provides:

• Model version tracking: Commit your Modelfiles and configuration for each client model version. Know exactly what changed between v1 and v2.
• Script management: Your data cleaning scripts, evaluation harnesses, and deployment scripts are versioned and backed up.
• Client-specific branches: Each client has their own branch or repository. Their configuration never contaminates another client's.
• Collaboration ready: When you eventually hire a contractor or employee, onboarding them to a codebase is faster than onboarding them to a collection of files on your laptop.

Private repositories: GitHub free plan includes unlimited private repos for individuals.

Total Monthly Overhead

At $1,000/month per retainer client (mid Tier 2), two clients cover the entire stack. Everything above two clients is gross margin.

What This Stack Lets You Do Alone

With this stack, one person can realistically manage:

• 6-8 active retainer clients
• 1-2 concurrent project engagements
• Monthly retraining cycles for all retainer clients
• Client communication via Loom + async
• Automated billing and reporting

Revenue ceiling: 8 retainer clients at $900/month average = $7,200/month MRR. Add 1-2 projects per quarter at $8,000-15,000 each, and a solo agency with this stack can generate $100,000-140,000/year with no full-time employees.

When to Hire Your First Person

Hire when: you are turning away qualified clients due to capacity constraints AND your revenue is consistently above $10,000/month. Hire for the work you least want to do (usually: client communication and data engineering), not the work you enjoy (usually: training and model evaluation).

---

Further Reading

• How to Start an AI Agency in 2026 — The full agency launch playbook
• AI Agency Retainer Model — Building recurring revenue from model maintenance
• Manage Multiple Fine-Tuned Models — Operational systems across clients
• Ertas Agency Plan Review — The platform behind the stack
• Recurring Revenue with Fine-Tuned Models — The business model