AI cost router for automation agencies
Use cheap inference for non-sensitive client workflows, fall back to premium or BYOK providers for sensitive and critical requests, and track usage per client.
OpenAI-compatible · n8n / Make / Dify / Telegram-ready · Client-level API keys · USDT / card billing
Agencies lose margin when LLM usage is invisible
Client usage is mixed together
One key, one bill — you cannot see which client burned the tokens.
Premium models where cheap would do
Bulk SEO, summaries, public FAQ and test runs rarely need expensive models.
One runaway workflow kills your margin
Without limits and alerts, a single client automation can outspend the retainer.
Sensitive and public calls routed the same
Public content can go cheap; private client data should use premium / BYOK.
One backend for all your client AI workflows
- ✓OpenAI-compatible endpoint — change one base URL
- ✓Separate API key per client
- ✓Usage tracked by client, model and day
- ✓Cheap mode for non-sensitive bulk workloads
- ✓USDT / card top-ups, no KYC, no lock-in
- ✓n8n, Make, Dify, Flowise, Telegram guides
- ✓Spend limits & alerts (rolling out)
- ✓Balanced fallback + BYOK modes (rolling out)
Items marked “rolling out” are set up by hand for founding agencies today.
Route each request by cost, sensitivity and reliability
Cheap is live today. Balanced and BYOK are rolling out for founding agencies.
Cheap
Available nowPublic, synthetic and non-sensitive workloads — SEO, content, product descriptions, public docs, summaries, tests and staging.
Best-effort availability on the decentralized inference network.
Balanced
Rolling outCheap first, fallback to a premium provider on timeout, error or latency threshold. For client-facing bots where cost matters but failures hurt UX.
In private rollout for founding agencies.
Premium / BYOK
Rolling outUse your own provider keys for sensitive clients while keeping the Mingles dashboard, limits and usage analytics.
In private rollout for founding agencies.
Know which client used what
Set hard limits before usage eats your retainer.
| Client | Tokens today | Cost today | Model mix | Error rate | Monthly limit | Alert |
|---|---|---|---|---|---|---|
| Client A — SEO bot | 4.2M | $3.10 | MiniMax 80% · Kimi 20% | 0.2% | $50 / mo | OK |
| Client B — support bot | 1.1M | $2.40 | Kimi 100% | 0.5% | $40 / mo | OK |
| Client C — content batch | 9.8M | $7.90 | MiniMax 100% | 0.1% | $60 / mo | 80% — near |
Illustrative preview. Per-client keys are live today; the dashboard and spend limits are rolling out for founding agencies.
Where agencies route cheaper
n8n / Make AI workflows
Change base URL. Keep the workflow. Track cost per client.
SEO / content automation
Route bulk generation to cheap models.
Lead enrichment & outreach
Process public data at a lower cost.
Telegram / WhatsApp bots
Cheap backend for community and FAQ bots.
RAG over public docs
Documentation bots without premium-model cost.
Testing & staging
Use cheap models before promoting to premium production.
Test it on one workflow before switching anything
- 1 Send us one existing workflow or bot (OpenAI / OpenRouter / Claude).
- 2 We map which calls are sensitive vs non-sensitive.
- 3 We route the safe calls through Mingles cheap tier.
- 4 We keep fallback / BYOK for the critical calls.
- 5 You get a cost, latency and quality comparison.
Send us one workflow. We’ll show what can safely route cheaper, where fallback is needed, and what your client-level cost would look like.
Request a migration test
We reply within one business day.
Agency pricing
Monthly base + usage. Pay by card or USDT. Cancel anytime.
Founding Agency
RecommendedFor your first few client workflows.
- ✓51M tokens / mo included (36M in + 15M out), then pay-as-you-go
- ✓3 client projects
- ✓Separate API key per client
- ✓Cheap mode for non-sensitive workloads
- ✓Spend tracking per client
- ✓USDT / card top-ups
- ✓We migrate 1 workflow with you
Agency Growth
For agencies running prod client bots.
- ✓189M tokens / mo included (135M in + 54M out), then pay-as-you-go
- ✓15 client projects
- ✓CSV export of usage
- ✓Webhook usage logs
- ✓Balanced fallback routing (rolling out)
- ✓We migrate 2 workflows with you
- ✓Priority Telegram support
Agency Pro
For studios at scale.
- ✓660M tokens / mo included (480M in + 180M out), then pay-as-you-go
- ✓50 client projects
- ✓BYOK routing for sensitive clients (rolling out)
- ✓Custom fallback policies
- ✓Higher rate limits
- ✓Monthly usage review
- ✓Early white-label dashboard
Not sure which tier? Send us one workflow and we’ll recommend one.
A note on sensitive data
Cheap mode is for public, synthetic and non-sensitive workloads. Do not send secrets, regulated data or private customer information through the decentralized cheap mode. Use Premium / BYOK mode for sensitive clients. See the routing guide for a data-sensitivity table.
Agency FAQ
▸Is this a replacement for OpenAI / OpenRouter?
Not always — it is a routing layer. Use cheap inference for non-sensitive bulk calls and premium / BYOK for critical calls. Many agencies run both.
▸Can we send private client data?
Do not send secrets or regulated / sensitive data through the cheap decentralized mode. Use Premium / BYOK mode for sensitive clients. See the routing guide for a data-sensitivity table.
▸What if the cheap tier is slow or unavailable?
Balanced mode (rolling out) falls back to a premium provider on timeout, error or latency. Today the cheap tier is best-effort; critical paths should wait for balanced/BYOK or use a premium key.
▸Does it work with n8n / Make / Dify / Flowise?
Anything OpenAI-compatible connects by changing the base URL and model ID. We provide migration guides and help you move your first workflow.
▸Can we track usage per client?
Yes — separate API keys per client today, with a per-client usage dashboard and spend limits rolling out for founding agencies (set up by hand in the meantime).
▸Can we pay in USDT?
Yes — USDT / USDC top-ups and card payments. No KYC — a monthly agency base plus usage, no long-term lock-in.