The routing layer for AI spend
Same quality, cheaper track.
Makas evaluates every request. When a lower-cost model can handle the task, it routes it there; when not, the request stays put. You see both prices.
Free for a limited time, for founding customers. No card, no payment, no commitment.
Detected data extraction, medium · rule: Everyday Balance
Works with the providers you already use. Each request stays with the provider you called.
OpenAI
GPT Astra · GPT Sol · GPT Terra · GPT Luna · GPT Pro · GPT Codex
Anthropic
Claude Fable · Claude Opus · Claude Sonnet · Claude Haiku
Gemini Argon · Gemini Pro · Gemini Flash · Gemini Flash-Lite
Sample request log
See what every request cost, and what it would have cost.
Eight example prompts, priced at today's catalog rates using the default routing rules. Each result belongs to that specific prompt, not every request of its type.
39%
saved across these eight examples
$0.01048 of $0.02718
| Request | Requested → routed | Tokens in / out | Would have cost | Cost | Saved |
|---|---|---|---|---|---|
| Support Chat Routed | claude-sonnet-5 → claude-sonnet-5-5 (modernized at no extra cost) | 700 / 150 | $0.00290 | $0.00290 | — |
| Contextual Q&A Routed | gpt-6-sol → gpt-6.1-terra | 1,500 / 180 | $0.00480 | $0.00240 | $0.00240 (50%) |
| Code review Routed | gpt-5.4 → gpt-6.1-terra | 600 / 250 | $0.00525 | $0.00185 | $0.00340 (65%) |
| Agent plan Stays | claude-opus-5-5 | 900 / 120 | $0.00600 | $0.00600 | — |
| Summarize Routed | gemini-3.8-flash → gemini-2.5-flash | 800 / 60 | $0.00083 | $0.00039 | $0.00044 (53%) |
| Translate Routed | gemini-3.1-pro-preview → gemini-2.5-flash | 120 / 130 | $0.00180 | $0.00036 | $0.00144 (80%) |
| Analyze Routed | gpt-6-sol → gpt-6.1-terra | 900 / 180 | $0.00360 | $0.00180 | $0.00180 (50%) |
| Extract Routed | gpt-6-sol → gpt-6.1-terra | 350 / 130 | $0.00200 | $0.00100 | $0.00100 (50%) |
Routing details for this Extract exampleHow the request above reached its model, step by step.
| |||||
| 8 sample requests | 7 routed (1 same price), 1 stays | $0.02718 | $0.01670 | $0.01048 (39%) | |
These are sample results, not a prediction for your workload. Routing depends on the prompt, requested model, and active rules. The examples use typical request sizes derived from public data; requests that stay on their original model are included in the totals.
Setup
Change the base URL. Add your Makas key. Nothing else.
Keep your SDK and your provider key.
- 01
Point your SDK at Makas
Set the base URL and send your Makas key in a Makas-Api-Key header. Your SDK and provider key stay as they are.
- 02
We classify and route
Each request is classified by task and complexity. When a lower-cost model from the same provider can handle it, it moves there. If not, it stays put or moves to a model at the same price.
- 03
See both prices
Every request is logged with what it cost and what it would have cost, and shows up in your request log as it happens.
- Streaming works as before.
- Your provider key passes through with each request. Makas uses it for that request and doesn't store it.
- New base URL, and your Makas key in a Makas-Api-Key header.
- Your provider key is unchanged. It passes through and Makas doesn't store it.
Base URLhttps://api.openai.com/v1 becomes https://eu.makas.ai/proxy/openai/v1
Routing & transparency
See how your request would route
Try the routing engine you’d get on signup. Inspect the model choice and estimated cost for your prompt; no model answer is generated.
Live routing preview
Start with an example
All numbers are estimates. Token counts are heuristic. Actual savings vary.
In the app
Everything here is in the app today.
Plans are marked on each feature. The Founding Customer plan includes all nine.
Routing decision
For every request: what was detected, which rule matched, and both prices.
All plans
Request log
Every request with its model, cost and saving, as it happens.
All plans
Cost dashboard
Spend and savings over time, with your top models.
All plans
Savings analytics
Where every saving comes from: by strategy and rule, by provider and model, with request activity and latency alongside.
All plans
Conversation-aware routing
A multi-turn conversation stays on the model it started on, so context and tone hold for the whole thread.
All plans
Model upgrades
Providers ship new models every few weeks. When a newer model costs the same or less, your requests move to it, and before a model is retired you can move off it. No model names to change in your code.
Growth and up
Routing rules
Your own rules run before the built-in cost strategies.
Custom rules on Growth and up
Spending caps and alerts
Monthly caps, with alerts at 50, 80, 90 and 100%.
Growth and up
Audit trail and roles
Admin, Developer and Viewer roles. Key, rule and cap changes are recorded.
Scale
Architecture
We sit in your request path, so we're built like it.
Here is what happens to each request, in order.
- 01
Prompt analysis
Each request is classified by task type, such as code, creative or analytical, and by complexity: simple, medium or complex. A small embedding model running inside Makas classifies by default, so classification sends the prompt nowhere else. Keyword matching is the fallback.
Prompt plaintext is not persisted during normal routing.
- 02
Routing decision
Your rules are evaluated first, then the defaults, each in priority order. Rules that can raise cost are off by default, and routing stays with the provider you called.
Each decision is logged with the model, the rule and both prices.
- 03
Security
Your provider key comes with each request and is used in memory to call the provider. We don't store it. Makas API keys are stored hashed. Sessions use JWTs signed with RS256.
Scale adds an audit trail and role-based access.
- 04
Resilience
Each provider has its own circuit breaker, so an outage fails fast instead of hanging. Non-streaming requests retry provider errors with backoff.
When a breaker is open, Makas fails the request at once instead of waiting on the provider.
(Coming soon) Semantic caching is the exception. On Growth and up, it sends the prompt to OpenAI to create an embedding, including for requests to Anthropic and Google, keeps a SHA-256 hash and an unencrypted embedding, and stores the response encrypted per-customer. Retention is enforced, and there is no separate opt-out. It is off during the Founding period.
Pricing
A flat monthly price. No commission on your savings.
Start on the Founding Customer plan. The paid plans below come later.
Founding Customer
Freefor a limited time
A special plan for our first customers. No card, no payment, no commitment. Start with one service's production traffic.
Included
- Full analytics, 30-day request log
- 10 custom routing rules
- Team of up to 5
- Spending caps and alerts
- Audit trail and roles
- Per-key routing and spending caps
- CSV export
Limits
3B tokens and 5 million requests a month, 200 requests a minute. Hard limits.
Need more? Contact us and we'll raise your limits.
When paid plans launch
- 50% off the first year of an annual paid plan started within the 14-day notice period
- 14 days to switch to a paid plan after announcement, before any charge
Not included
Semantic caching, which sends prompts to OpenAI for embedding, is off during the Founding period.
Paid plans · coming later. Plan contents and prices may change before launch.
- Planned
Starter
3B tokens/month
$49 /mo
or $470/yr, billed annually
Founders pay $235 for the first year (50% off)
For one developer routing production traffic.
Included in all tiers:
- Dashboard & Trends
- Advanced Analytics
- Default Optimization Strategies
- Conversation-Aware Routing
- Cache Optimization
- Data Export
- Data Isolation
- Planned
Growth
15B tokens/month
$199 /mo
or $1,910/yr, billed annually
Founders pay $955 for the first year (50% off)
For teams that want caching, spending caps and their own rules.
Everything in Starter, plus:
- Automatic Model Upgrades
- Custom Routing Rules
- Isolated Semantic Caching
- Monthly Spending Limits
- Budget Alerts
- Team Management
- Planned
Scale
50B tokens/month
$499 /mo
or $4,790/yr, billed annually
Founders pay $2,395 for the first year (50% off)
For high-volume teams that need routing controls and audit visibility.
Everything in Growth, plus:
- Per-Key Routing Control
- Per Key Spending Limits
- Role-Based Access Management
- Audit Trail
Need more than 50B tokens a month? Talk to us
Mind the spread.
Run the Playground on one of your prompts, or sign up free and route your own traffic.
No card, no payment, no commitment.