Pricing & Credits

The SaaS edition uses unified Credits for billing, abstracting multi-vendor pricing. Private has no platform billing.

Why Credits

Different model vendors meter differently:

OpenAI: per thousand tokens / per image
Anthropic: per thousand tokens (different in/out price)
Some Embedding: per thousand tokens
Some Search APIs: per call
Some Reranking: per document pair

The business can't manage cost with one number. Evose abstracts these into Credits:

1 Credit = a certain number of tokens / calls / capacity (Evose dynamically calibrates internally per contract)

Every cost you see is in Credits. At month-end, you look at one number.

Consumption Dimensions

Dimension	Examples
Model calls	LLM tokens (in + out) · Embedding tokens · Reranking pairs · Image generation · Audio · Video
Tool calls	Per-call billing for EvoTool marketplace tools · Auto-conversion for paid third-party APIs
Knowledge base	Storage capacity (monthly)
Data source	Storage capacity (monthly)
Observability	Trace retention 30 days free; longer by capacity

Attribution

Every Credit consumption is attributed across three layers:

1. Workspace    Which team produced it
2. Resource     Which Agent / Workflow / KB produced it
3. User         Which end user produced it

→ Observability

Budgets and Alerts

In [Org · Settings]:

Monthly budget: 10000 Credit
Alert thresholds:
  - 50% → notify admin
  - 80% → notify admin + workspace owner
  - 100% → notify + pause non-critical workspaces (optional)

100% does not auto-stop everything

By default it only notifies. Forced shutdown (e.g. auto-pause) is an explicit toggle to avoid accidental disruption.

A Sample Estimate

Assume:

10,000 customer-service conversations per month
Average per turn: input 200 tokens / output 300 tokens / one KB retrieval
Model: Claude Sonnet 4.6

Rough math:

Model tokens:  10,000 × (200 + 300) = 5,000,000 tokens
              5,000 thousand tokens × unit X = ~ Y Credits

Vector retrieval:  10,000 × ~30-token query × unit = small
                     TopK=5 chunks each → also small

KB storage:    100 MB × unit = small

Total:         Y + retrieval + storage ≈ monthly cost

Credits estimator

Adjust the sliders for a real-time monthly estimate

Monthly conversations10,000

Avg input tokens / call200

Avg output tokens / call300

Knowledge base size (MB)100

Tool calls / month2,000

Estimated monthly credits

20.0Credits

Estimate based on platform default rates. Actual cost depends on your contract and is reflected in the observability panel.

Estimation basis

Above prices are illustrative. Actual cost is per contract + the current vendor's pricing. See real-time attribution in Observability.

Billing View Under Private

Private has no platform commission, but you can configure internal unit prices in Evose for cost attribution:

[Model] gpt-4-turbo
  input  cost: 0.01  / 1k tokens
  output cost: 0.03  / 1k tokens
 
[Reports] Allocate total cost across workspace / resource / user

Anti-Patterns

Defaulting the entire company to the most expensive model — pick by scenario; CS / FAQ work well with Claude Haiku 4.5
Going live without monitoring — a runaway Workflow can burn a month's budget in a day
Stuffing all data into the knowledge base — bad for both storage costs and retrieval quality

Next Steps

See real-time consumption → Observability
Smooth migration → SaaS vs Private

Pricing & Credits

On this page