Everything you need. Nothing you don't.

See your total AI spend update as calls happen. Your cost counter ticks in real time.

Running GPT-4o and Claude Sonnet side-by-side? See which one is actually cheaper for your workload, factoring in real token usage — not benchmark numbers.

OpenAI, Anthropic, and Google costs aggregated in one view. Compare provider spend at a glance.

7-day, 30-day, and 90-day cost trend charts. See whether your spend is growing proportionally to your traffic, or running away from it.

Costara flags unusual spikes in spend so you find out about runaway jobs before your bill does.

Tag each LLM call with the product feature that triggered it (`customer-support-chat`, `document-summarizer`, `weekly-digest`). Costara groups and aggregates by tag automatically.

Side-by-side cost charts for your most expensive features. Instantly see which ones are scaling linearly and which are out of control.

A ranked list of your features by total cost and cost-per-call. Identify which features are bleeding your budget and which are delivering value.

Track `development`, `staging`, and `production` separately. Stop letting dev experiments pollute your production cost data.

`costara.init(api_key=..., project=..., environment=...)`. Works in any Python app, framework, or script.

Call `costara.track(...)` after any LLM API call. Full control over what gets tracked and how it's tagged. Works with OpenAI, Anthropic, Google, and any other provider.

Drop in `costara.patch_openai()` and all OpenAI calls are tracked automatically, zero per-call changes needed. Ideal for getting started quickly.

SDK calls are batched and sent asynchronously. No added latency on your critical path. Fully compatible with `asyncio`, FastAPI, and async Django.

Disable all SDK tracking with a single environment variable (`COSTARA_DISABLED=1`) without changing code. Useful for local development or incident response.

Configure API key, project, and environment via environment variables. No need to change code between environments.

`costara.estimate_cost(provider, model, usage)` returns the exact cost using up-to-date provider pricing tables.

Set a monthly spending cap per project. Track your burn rate against the cap in real time.

Configure daily, weekly, or monthly budget periods depending on how your team tracks spend.

Receive alerts at 80% and 100% of your budget. Customize thresholds to suit your team's workflow.

Optionally enable hard-stop mode: Costara returns a flag your SDK can check before making an LLM call, allowing you to gate calls when you're over budget.

Based on your current daily burn rate, Costara projects whether you'll hit your monthly budget and when.

Invite teammates to your Costara organization. Engineers, PMs, and finance can each see the data relevant to them.

Four roles with distinct permissions: Owner — Full access, billing, and settings Admin — Manage members, projects, and budgets Member — View all data, manage own projects Viewer — Read-only access to dashboards and reports

Invite teammates by email. They join your organization directly without needing a separate account first.

Members of multiple organizations can switch between them from the dashboard with a single click.

Separate projects for each product, service, or client. Costs and budgets are tracked independently.

Receive budget threshold alerts and weekly digest reports directly in your Slack workspace.

SMTP-based alerts for teams not on Slack.

OpenAI (auto-instrument + explicit), Anthropic (explicit), Google (explicit). More providers in the roadmap.

POST to any endpoint when a budget threshold is crossed. Build your own automations.

Query your cost data programmatically. Build internal dashboards or pipe data into Grafana, Datadog, or your own data warehouse.

Costara records metadata only: token counts, model name, provider, latency, cost, and feature tag. Your prompts and completions are never sent to Costara's servers. This is a structural constraint of the SDK — not just policy.

Costara API keys are scoped per organization, project, and environment. Your Costara key is entirely separate from your LLM provider keys. Costara never touches your OpenAI or Anthropic credentials.

All data is isolated per organization using Supabase RLS policies enforced at the database layer. One team cannot access another team's data regardless of application-level checks.

All data is encrypted at rest (AES-256) and in transit (TLS 1.2+). Your data is never transmitted in plain text.