What's new in Framewren

Gemini 1.5 Pro and Flash support

Framewren now tracks cost and latency for Gemini 1.5 Pro and Gemini 1.5 Flash. Character-based pricing is normalized to a per-token equivalent in your dashboard so you can compare directly with OpenAI and Anthropic spending.

• Cost per request in normalized token units
• P50/P95/P99 latency separated from network overhead
• Side-by-side comparison in the provider breakdown view

Alert threshold builder redesign

The alert configuration UI has been rebuilt from scratch. You can now define multi-condition thresholds — for example, trigger when P95 latency exceeds 3.5 seconds AND cost exceeds $0.002 per request over a 15-minute window. Previous alerts are migrated automatically.

Project-level cost attribution tags

Attach arbitrary metadata tags to instrumented calls — feature name, customer tier, experiment ID, environment. The cost dashboard now lets you group and filter spending by any tag, so you can see exactly which product surface or A/B variant is driving LLM costs.

SDK v0.9 — latency precision and streaming fix

Fixed a measurement artifact where streaming responses reported first-token latency as total-response latency in certain async configurations. Latency is now split into TTFT (time-to-first-token) and total response duration, surfaced separately in the dashboard.

Monthly budget caps with Slack and email notifications

Set a monthly spend limit per project. Framewren sends a Slack or email notification at 75%, 90%, and 100% of budget. No hard cutoff — just visibility so your team can act before overage surprises.

Framewren launch — public beta

Framewren is live. OpenAI, Anthropic, and Mistral integrations are available at launch. The Python SDK is stable and the Node.js SDK is in beta. Sign up for free — no credit card required.