If you have evaluated AI agent platforms recently, you have probably noticed the pricing pages are designed to be confusing. There is a platform fee, plus credits, plus per-token charges, plus overages. You sign up for $19/month and end up paying $300 because your agents were chatty.
This is the problem BYOK solves, and it is the core of how Swrly handles pricing.
The Hidden Cost Problem
Most AI platforms bundle two separate things into one bill: the orchestration layer (workflow management, integrations, UI) and the AI compute (LLM inference, token usage). Bundling them together lets platforms charge unpredictable per-token markups that scale with your usage.
Here is what the landscape looks like today:
Relevance AI charges $19/month for the Pro plan, but that includes a limited number of credits. Each agent run consumes credits based on token usage. Complex workflows with multiple agent steps can burn through credits in days. Overages are billed per-credit, and the cost per credit varies by model.
CrewAI starts at $200/month for the Enterprise plan. It includes "unlimited" agents, but inference costs are billed separately based on the LLM provider and token volume. The total monthly cost depends entirely on how many tokens your agents consume.
LangSmith/LangChain charges per trace for observability, and you still pay your LLM provider separately. The traces add up — a busy team running thousands of workflows can see observability costs alone exceed $100/month.
In every case, the total cost is unpredictable until you get the bill.
Swrly's BYOK Model
Swrly separates the two costs completely:
- AI compute: You pay Anthropic directly for your Claude Code subscription. $20/month for Pro. That is between you and Anthropic — Swrly never touches it.
- Orchestration: You pay Swrly for the platform. Free for individuals, $49/month for Pro, $99/month for Teams. Fixed price, no per-token charges, no credit system.
The mechanism is straightforward. In Swrly's Settings page, you paste your Claude Code session token. Swrly encrypts it using AES-256-GCM (the same encryption standard used by banks and government systems) and stores the encrypted token in the database. When a workflow runs, the token is decrypted ephemerally in memory, used for that execution, and discarded. It is never written to logs, never stored in plaintext, and never accessible to Swrly staff.
Your key, your costs, your control.
The Cost Comparison
Here is what a team of 5 engineers running approximately 1,000 agent workflow executions per month can expect to pay:
| | Swrly + Claude Code | Relevance AI | CrewAI |
|
Top comments (0)