Pick the agent you need, the systems it connects to and how much work it will handle. Get a build cost, a monthly running cost (LLM API, hosting, maintenance) and a payback period — with every assumption shown below.
Typical scope for each agent type, priced with the same model as the calculator. Free to use — embed the chart on your site with a link back to this page.
| Agent type | Typical scope | Build cost | Build time | Running cost / month* |
|---|---|---|---|---|
| FAQ / knowledge-base chatbot | 1 integration, small knowledge base, website widget | $1,100 – $1,800 | 2 – 4 weeks | ≈ $280 |
| Workflow automation agent | 3 integrations, email intake, human approval step | $3,100 – $5,150 | 4 – 8 weeks | ≈ $530 |
| Sales / research agent | 3 integrations, Slack, long-term memory | $4,150 – $6,850 | 4 – 10 weeks | ≈ $1,260 |
| Customer support agent | 2 integrations, 100–1,000 docs, web + email, escalation, dashboard | $5,450 – $8,850 | 5 – 11 weeks | ≈ $840 |
| Multi-agent system | 5 integrations, 100–1,000 docs, 2 channels, all extras | $12,250 – $20,500 | 10 – 23 weeks | ≈ $3,060 |
*5,000 conversations or tasks per month on Claude Sonnet 5.5, including hosting and maintenance. Build time assumes one engineer; larger teams finish sooner.
Built on BinaryBits' experience delivering AI agent and automation projects, plus the published list prices of the major LLM APIs. Every figure on this page comes from the assumptions below — nothing is hidden.
We estimate a low and high number of engineering hours for each part of the project, add 15% for project management and QA, and multiply by our hourly rate.
| Component | Hours | What it covers |
|---|---|---|
| Core build — FAQ / knowledge-base chatbot | 50 – 80 | Discovery, prompt & tool design, agent logic, evaluation set, deployment |
| Core build — Customer support agent | 90 – 140 | Discovery, prompt & tool design, agent logic, evaluation set, deployment |
| Core build — Workflow automation agent | 80 – 130 | Discovery, prompt & tool design, agent logic, evaluation set, deployment |
| Core build — Sales / research agent | 100 – 160 | Discovery, prompt & tool design, agent logic, evaluation set, deployment |
| Core build — Multi-agent system | 200 – 320 | Discovery, prompt & tool design, agent logic, evaluation set, deployment |
| Each system integration | 10 – 18 | Auth, API client, error handling and tests for one external system |
| Knowledge base — Medium (100 – 1,000 documents) | 15 – 25 | Ingestion pipeline, chunking, embeddings, retrieval tuning (small KBs are in the core build) |
| Knowledge base — Large (1,000+ docs, mixed formats) | 35 – 60 | Ingestion pipeline, chunking, embeddings, retrieval tuning (small KBs are in the core build) |
| Each extra channel | 8 – 14 | Connecting the agent to another surface (first channel included) |
| Human approval / escalation step | 12 – 20 | Approval queue, hand-off to a human, notifications |
| Admin dashboard & analytics | 30 – 50 | Conversation logs, usage & cost metrics, settings UI |
| Long-term memory & personalisation | 20 – 35 | Per-user memory store, recall and privacy controls |
| Security & compliance hardening (PII redaction, audit logs, SSO) | 20 – 40 | PII redaction, audit trail, SSO, access controls |
| Project management & QA | +15% | Applied to the total above |
An agent rarely makes a single model call per task: it plans, calls tools, reads results and replies, and each call re-sends the conversation so far. Our defaults reflect typical production agents we build:
| Agent type | LLM calls / task | Input tokens / call | Output tokens / call |
|---|---|---|---|
| FAQ / knowledge-base chatbot | 4 | 2,500 | 350 |
| Customer support agent | 6 | 3,500 | 400 |
| Workflow automation agent | 5 | 3,000 | 600 |
| Sales / research agent | 12 | 5,000 | 700 |
| Multi-agent system | 25 | 6,000 | 800 |
| Model | Tier | Input | Output |
|---|---|---|---|
| Claude Haiku 5.5 | Budget | $0.10 | $0.50 |
| Claude Sonnet 5.5 | Mid | $2.00 | $10.00 |
| Claude Opus 5.5 | Frontier | $4.00 | $20.00 |
| GPT-6 Luna | Budget | $0.10 | $0.50 |
| GPT-6.1 Sol | Mid | $2.00 | $10.00 |
| GPT-6 Astra | Frontier | $10.00 | $50.00 |
Sources: Anthropic API pricing · OpenAI API pricing. We use list prices without prompt-caching or batch discounts, so real bills are usually lower — caching alone often cuts input costs by 50% or more for agents with long, repeated system prompts.
We only count tasks the agent completes end-to-end. Time saved on partially assisted tasks, faster response times and 24/7 coverage are real benefits but are left out to keep the estimate conservative.
This is a planning estimate, not a quote. Projects with unusual APIs, strict compliance (HIPAA, SOC 2) or poor data quality can exceed the high end.
Hourly rates reflect BinaryBits' pricing. Agencies in the US or Western Europe typically bill several times more per hour, so scale the build cost accordingly if you are comparing vendors.
BinaryBits. AI Agent Cost Calculator: Build & Running Costs. https://binarybits.co/ai-agent-cost-calculator/
Tell us what you want to automate. We'll confirm scope, pick the right model for your budget and send you a fixed price.