Cost attribution
Spend broken down by person, team, project, model and provider — down to the token.
Tokkara · AI cost intelligence by EMOR
See everydollar yourAI burns.
Tokkara tracks AI spend across AWS Bedrock, OpenAI and Anthropic Claude in one dashboard — cost per person, project and model, with the fixes ranked by how much they save. Read-only. No SDK, no proxy, no code changes.
- Month to date
- $15,690
- Projected month end
- $21,400
- Daily run rate
- $712
- Savings found
- $3,750 / mo
Spend by model
- Claude Opus58%
- GPT-524%
- Nova Lite11%
- Other7%
One view across
- AWS Bedrock
- OpenAI
- Anthropic Claude
01 — The problem
The bill grows. The answer doesn’t.
Sound familiar? 01
The AI bill doubled and nobody can say why.
Sound familiar? 02
Finance wants spend by team. You have three consoles and a spreadsheet.
Sound familiar? 03
You suspect the biggest model is used for everything. You can’t prove it.
02 — Savings Engine
Fixes ranked by dollars saved.
Tokkara doesn’t stop at a report. It prices every change — smaller models, prompt caching, batch jobs, output caps — counts overlapping savings once, and tracks each fix until the saving is proven.
| Action | Effort | Saving / mo |
|---|---|---|
| Route summarisation traffic from Opus-class to a smaller model | Low | $1,840 |
| Turn on prompt caching for the support assistant’s system prompt | Low | $960 |
| Move nightly document jobs to batch pricing | Medium | $710 |
| Cap output tokens on internal chat | None | $240 |
| Combined, each saving counted once | $3,750 | |
03 — How it works
It reads what your providers already log.
01
Connect read-only
Add a cross-account IAM role for Bedrock and admin keys for OpenAI and Anthropic. Nothing sits in the request path.
02
We sync and price
Usage is pulled on a schedule, matched to people and projects, and priced from one maintained rate table.
03
You see one bill
Every provider, model and person in a single dashboard, with run rate and a month-end forecast.
04
You act on ranked fixes
The Savings Engine ranks changes by dollars saved, then tracks the rollout and proves the saving.
- AWS Bedrock
- CloudTrail model-invocation events, via a read-only cross-account role
- OpenAI
- Organization usage and costs API, with an admin key
- Anthropic Claude
- Organization usage reports, with an admin key
04 — Capabilities
It doesn’t just report. It prices the fix.
Savings Engine
Model right-sizing, caching, batch and token-cap recommendations, each priced per month and counted once.
Proven savings
Track each recommendation from proposed to rolled out, and see what it actually saved.
Forecast & alerts
Run rate, month-over-month change, month-end projection, budget thresholds and spike alerts.
Security findings
Exposed and long-lived access keys, grouped by owner, so leaks are caught before they are billed.
Reports
Showback-ready spend and savings reports for finance and leadership.
Teams & roles
Multi-tenant workspaces with owner, admin and viewer roles, plus Google sign-in.
Privacy first
Metadata and cost only. Prompt content is never stored by default.
05 — Who it’s for
One answer for every team.
Platform & IT
Find which workloads drive cost and fix them in order of return.
Finance
Spend by team and project, a forecast, and numbers that match the invoice.
Security
Know which keys are exposed and who owns them.
Engineering
See the cost of each model choice before it ships.
06 — Get started
Start with a free assessment.
We connect read-only to your AI providers and hand you a clear picture of where the money goes.
- Spend broken down by person, project, model and provider
- Your run rate and projected month-end bill
- A ranked list of savings, priced per month
- Exposed keys and other security findings
07 — FAQ
Questions, answered.
What is Tokkara?
Tokkara is an AI spend analyzer built by EMOR (Emortech). It brings your AWS Bedrock, OpenAI and Anthropic Claude usage into one dashboard, attributes cost to people, projects and models, and ranks the changes that would cut your bill.
Which AI providers does Tokkara support?
AWS Bedrock, OpenAI and Anthropic Claude, including API usage and subscription seats such as ChatGPT Team or Claude Team. Every provider is normalised into the same view, so you can compare models across clouds.
Do I need to change my code or route traffic through a proxy?
No. Tokkara reads the usage data your providers already record — CloudTrail for Bedrock and the admin usage APIs for OpenAI and Anthropic. It never sits in the request path, so there is no added latency and nothing to deploy.
How does Tokkara attribute AI cost to individual users?
It maps the identity on each request — the IAM user or role for Bedrock, the user, project or API key for OpenAI and Anthropic — to a person and team, then prices every request from a single rate table.
How does Tokkara reduce AI costs?
The Savings Engine looks for over-sized models, cacheable prompts, batchable jobs and uncapped output, prices each fix per month, and ranks them by return. Overlapping savings are counted once, and rolled-out fixes are tracked to the saving they actually delivered.
Does Tokkara store my prompts?
No. By default it stores metadata and cost only — model, tokens, identity and time. Prompt content is never retained unless you opt in.
How do I get started?
Book a free cost assessment. We connect read-only to your providers, and you get a breakdown of where your AI spend goes and a ranked list of savings.