AI Token Usage Dashboard

Goal

Provide one place to understand AI token consumption across OpenAI, Azure OpenAI, AWS Bedrock, Google Vertex AI, Anthropic AI, Oracle AI, GitHub Copilot, Meta AI, xAI, Perplexity AI, Mistral AI, Cohere, DeepSeek, Manus AI, Sarvam AI, Krutrim, and Cursor — by provider, model, date, and application when AISP metadata is available.

Need

Each cloud provider exposes usage differently. This dashboard normalizes metrics into a common view so platform, FinOps, and engineering teams can compare consumption, validate integrations, and drill into request-level details without switching consoles.

Demo mode — works immediately with synthetic data (no credentials).
Live mode — queries configured providers server-side; one failing provider does not block others.
Request details — click a Requests count to see per-request rows; AppName appears when AISP tags or metadata provide it.

Deployment

Copy .env.example to .env, install dependencies, and run:

Local: uvicorn app.main:app --reload --port 8000
Docker: docker compose up --build or ./br.sh start

Set environment variables for each provider you want in live mode. Providers appear as not configured until their required variables are set. API docs: /docs · Prompt Runner: /prompt-runner

Configure

Enable a provider by setting its usage variables in .env, restarting the service, then switching the dashboard to Live mode. Full variable lists live in .env.example and README.md.

GitHub Copilot

Ensure Copilot Business/Enterprise is active and the Copilot usage metrics policy is enabled for your org or enterprise.
Create a PAT with metrics read access: classic read:org (org) or manage_billing:copilot / read:enterprise (enterprise); or fine-grained View Organization Copilot Metrics (or enterprise equivalent).
Set in .env:
DEMO_MODE=false
GITHUB_COPILOT_TOKEN=ghp_… (or GITHUB_TOKEN)
GITHUB_COPILOT_ORG=your-org or GITHUB_COPILOT_ENTERPRISE=your-enterprise
Optional: GITHUB_COPILOT_REGION, GITHUB_COPILOT_CREDIT_USD, GITHUB_COPILOT_FETCH_CONCURRENCY, GITHUB_API_BASE_URL.
Restart, open Live mode, filter Provider → GitHub Copilot, and confirm Integration status is connected.

Prompt Runner (optional): set GITHUB_COPILOT_RUN_MODEL (default gpt-4.1), choose GitHub Copilot on /prompt-runner, and run in live mode. Live completions create billable Copilot usage.

Meta AI

Create an API key at dev.meta.ai (Model API) or llama.developer.meta.com (Llama API).
Set in .env:
DEMO_MODE=false
META_AI_API_KEY=… (or MODEL_API_KEY / LLAMA_API_KEY)
META_AI_BASE_URL=https://api.meta.ai (or https://api.llama.com/compat for Llama)
Optional: META_AI_USAGE_PATH for an OpenAI-shaped usage endpoint; META_AI_INPUT_USD_PER_MTOK / META_AI_OUTPUT_USD_PER_MTOK for cost estimates.
Restart, open Live mode, filter Provider → Meta AI, and confirm Integration status is connected.

Prompt Runner (optional): set META_AI_RUN_MODEL (default muse-spark-1.1), choose Meta AI on /prompt-runner, and run in live mode.

xAI · Perplexity AI · Mistral AI · Cohere · DeepSeek

These AISPs use OpenAI-compatible chat and usage endpoints. Set the API key in .env, optionally override base URL / usage path / default run model, restart, and switch to Live mode.

xAI — XAI_API_KEY, base https://api.x.ai, model grok-4.5. Docs: docs.x.ai
Perplexity AI — PERPLEXITY_API_KEY, base https://api.perplexity.ai, model sonar-pro. Docs: docs.perplexity.ai
Mistral AI — MISTRAL_API_KEY, base https://api.mistral.ai. Docs: docs.mistral.ai
Cohere — COHERE_API_KEY, base https://api.cohere.ai (chat via /compatibility/v1). Docs: docs.cohere.com
DeepSeek — DEEPSEEK_API_KEY, base https://api.deepseek.com. Docs: api-docs.deepseek.com

If historical usage is not exposed on the account, set the provider *_USAGE_PATH to a gateway that exports OpenAI-style organization usage, or use demo mode.

Integrations

Official dashboards and docs for each AISP. Open these to create keys, enable metrics, and verify API access before switching this app to Live mode.

OpenAIUsage dashboard Usage API Admin keys
Azure OpenAIAzure OpenAI docs Azure Monitor metrics Azure portal
AWS BedrockBedrock docs API keys CloudWatch metrics Bedrock console
Google Vertex AIVertex AI docs Cloud Monitoring Google Cloud console
Anthropic AIAnthropic docs Usage & Cost Admin API Anthropic console
Oracle AIOCI Generative AI OCI Monitoring Oracle Cloud
GitHub CopilotUsage metrics API View usage metrics Personal access tokens
Meta AIMeta Model API Model API docs Llama API Llama quickstart
xAIxAI docs xAI console API reference
Perplexity AIPerplexity docs API settings Chat completions
Mistral AIMistral docs Mistral console API reference
CohereCohere docs Cohere dashboard Chat API
DeepSeekDeepSeek docs DeepSeek platform Chat completions
Manus AIManus API docs Authentication Team usage stats manus.im
Sarvam AIAPI docs Authentication sarvam.ai
KrutrimCloud docs Krutrim Cloud olakrutrim.com
CursorAdmin API Cloud Agents API keys

Manus AI

Create an API key in Manus API Integration settings.
Set in .env:
DEMO_MODE=false
MANUS_API_KEY=…
MANUS_BASE_URL=https://api.manus.ai
Optional: MANUS_CREDIT_USD, MANUS_TOKENS_PER_CREDIT, MANUS_RUN_MODEL.
Restart, open Live mode, filter Provider → Manus AI. Team keys use daily credit stats; personal keys fall back to session usage history.

Prompt Runner (optional): Manus creates an agent task and polls until completion — live runs consume credits.

Sarvam AI

Create an API subscription key in the Sarvam console.
Set in .env:
DEMO_MODE=false
SARVAM_API_KEY=…
SARVAM_BASE_URL=https://api.sarvam.ai
Optional: SARVAM_USAGE_PATH, SARVAM_RUN_MODEL (default sarvam-30b).
Restart, open Live mode, filter Provider → Sarvam AI. Auth uses the api-subscription-key header.

Prompt Runner (optional): set SARVAM_RUN_MODEL, choose Sarvam AI on /prompt-runner, and run in live mode.

Krutrim

Create an API key in Krutrim Cloud.
Set in .env:
DEMO_MODE=false
KRUTRIM_API_KEY=…
KRUTRIM_BASE_URL=https://cloud.olakrutrim.com
Optional: KRUTRIM_USAGE_PATH, KRUTRIM_RUN_MODEL (default krutrim-2).
Restart, open Live mode, filter Provider → Krutrim.

Prompt Runner (optional): set KRUTRIM_RUN_MODEL, choose Krutrim on /prompt-runner, and run in live mode.

Cursor

Create an Enterprise team Admin API key at Cursor Dashboard → API Keys.
Set in .env:
DEMO_MODE=false
CURSOR_API_KEY=…
CURSOR_BASE_URL=https://api.cursor.com
Optional: CURSOR_REGION, CURSOR_USAGE_PAGE_SIZE.
Restart, open Live mode, filter Provider → Cursor. Usage comes from Admin API /teams/filtered-usage-events.

Prompt Runner (optional): set CURSOR_REPO_URL (GitHub repo), CURSOR_RUN_MODEL (default composer-2), choose Cursor on /prompt-runner, and run in live mode (Cloud Agents API).

Permissions

Credentials are used only on the server — never in the browser. Grant least privilege to the identity running this service. Use the Integrations links above to open each provider console.

Provider	Usage dashboard	Prompt Runner (live)
OpenAI	Organization Admin API key (`OPENAI_ADMIN_KEY`) with access to `GET /v1/organization/usage/completions`	`OPENAI_API_KEY` with permission to call the configured model
Azure OpenAI	`AZURE_OPENAI_RESOURCE_ID` + `DefaultAzureCredential` with Monitoring Reader on the Cognitive Services account	`AZURE_OPENAI_ENDPOINT`, API key, and deployment with chat completion access
AWS Bedrock	`AWS_REGION` + IAM access keys for CloudWatch (`AWS_ACCESS_KEY_ID` / `AWS_SECRET_ACCESS_KEY`); Bedrock API keys cannot read metrics	`AWS_BEDROCK_API_KEY` (or `AWS_BEARER_TOKEN_BEDROCK`) + `AWS_BEDROCK_MODEL_ID` for Converse
Google Vertex AI	`GCP_PROJECT_ID` + ADC with Monitoring Viewer (`monitoring.timeSeries.list`) and configured metric types	Vertex AI User (`roles/aiplatform.user`) to call the configured model
Anthropic AI	`ANTHROPIC_ADMIN_KEY` with access to `GET /v1/organizations/usage_report/messages`	`ANTHROPIC_API_KEY` with permission to call the configured model
Oracle AI	`OCI_COMPARTMENT_ID`, `OCI_REGION`, OCI config profile, and Monitoring read on configured MQL queries	OCI Generative AI Inference permission for `OCI_GENAI_MODEL_ID`
GitHub Copilot	`GITHUB_COPILOT_TOKEN` (or `GITHUB_TOKEN`) + `GITHUB_COPILOT_ORG` or `GITHUB_COPILOT_ENTERPRISE`; PAT needs Copilot usage metrics read (`read:org` or fine-grained View Organization Copilot Metrics). Metrics policy must be enabled.	Same token with Copilot chat access; set `GITHUB_COPILOT_RUN_MODEL` (default `gpt-4.1`)
Meta AI	`META_AI_API_KEY` (or `MODEL_API_KEY` / `LLAMA_API_KEY`) + `META_AI_BASE_URL`; OpenAI-compatible organization usage path (auto-discovered or `META_AI_USAGE_PATH`)	Same key for `POST /v1/chat/completions`; set `META_AI_RUN_MODEL` (default `muse-spark-1.1`)
xAI	`XAI_API_KEY` + `XAI_BASE_URL` (default `https://api.x.ai`); OpenAI-compatible usage path or `XAI_USAGE_PATH`	Same key for chat completions; `XAI_RUN_MODEL` (default `grok-4.5`)
Perplexity AI	`PERPLEXITY_API_KEY` + `PERPLEXITY_BASE_URL`; OpenAI-compatible usage or `PERPLEXITY_USAGE_PATH`	`PERPLEXITY_RUN_MODEL` (default `sonar-pro`)
Mistral AI	`MISTRAL_API_KEY` + `MISTRAL_BASE_URL`; OpenAI-compatible usage or `MISTRAL_USAGE_PATH`	`MISTRAL_RUN_MODEL` (default `mistral-small-latest`)
Cohere	`COHERE_API_KEY` + `COHERE_BASE_URL`; OpenAI-compatible usage or `COHERE_USAGE_PATH`	`COHERE_RUN_MODEL` (default `command-r-plus`)
DeepSeek	`DEEPSEEK_API_KEY` + `DEEPSEEK_BASE_URL`; OpenAI-compatible usage or `DEEPSEEK_USAGE_PATH`	`DEEPSEEK_RUN_MODEL` (default `deepseek-chat`)
Manus AI	`MANUS_API_KEY` + `MANUS_BASE_URL` (default `https://api.manus.ai`); uses `/v2/usage.teamStatistic` or `/v2/usage.list` (credits mapped to tokens)	Same key for `/v2/task.create`; `MANUS_RUN_MODEL` / `MANUS_AGENT_PROFILE`
Cursor	Enterprise Admin API key (`CURSOR_API_KEY` or `CURSOR_ADMIN_KEY`) with access to `POST /teams/filtered-usage-events`	`CURSOR_API_KEY` + `CURSOR_REPO_URL` for Cloud Agents (`POST /v1/agents`); set `CURSOR_RUN_MODEL` (default `composer-2`)

See the project README.md for full IAM policy examples, GitHub Copilot and Meta AI setup steps, and environment variable reference.

Usage

Select a time range with presets or custom start/end dates, then click Refresh.
Toggle Demo mode or Live mode with the button in the header.
Filter by Origin (Global or country abbreviation) in the header to focus on AISPs from one country.
Filter by Provider, change Group by (None, Date, AISP, Model, Region, AppName), and pick a Metric for table columns or chart values.
Switch between Table and Chart views; use chart type icons for line, bar, and pie charts.
When grouped by Date with a line chart, use Subgroup to split series by AISP or model.
Click column headers to sort; hover headers for metric definitions.
Click any Requests value to open request-level details.
Review Provider distribution (with country-of-origin flags) and Integration status from the settings icon in the header.

Estimated cost in demo mode is illustrative. For production billing, use provider invoices or a versioned price catalog.

Usage details