Catch a runaway LLM bill the same afternoon. Not at month-end.

Use them. They're a backstop, not an alarm. Each provider's limit covers only that provider, and it's either a hard cap that fails your requests when hit or an email at a fixed percentage of a number you set months ago. None of them knows what your normal Tuesday looks like. PulseMeter gives you one budget across every provider, spike detection against your own baseline, and warnings hours before a cap would hit, sent to Slack and email so a human decides before anything breaks.

Helicone, Langfuse, and Portkey are developer observability tools: they sit in your request path as a proxy or need an SDK in your code, and cost is one panel among many. LLMeter is closer to us. It also reads spend from read-only keys with no proxy, but it is a cost dashboard you open. PulseMeter is built as an alarm: hourly detection reconciled nightly against each provider's official report, spike detection against your own weekday baseline, month-end projections, connection-health checks, and coverage for seat-based plans like Claude Enterprise and for Amazon Bedrock. Most days you won't open it at all. That's the point.

No. We only read usage and cost metrics from your provider's reporting APIs: token counts and dollar amounts by day and model — and, for team plans like Claude Enterprise, per-person usage, cost, and activity counts (messages, sessions, commits). We never see prompts, responses, or the content of anyone's conversations or code. See our security page.

Not for API providers. We only call your provider's usage and cost reporting endpoints, which return aggregates of spend that already happened. They never run a model, so they consume zero tokens and add nothing to your invoice. The one exception is Amazon Bedrock, where AWS bills your account $0.01 per Cost Explorer API request — PulseMeter makes about two per night, roughly $0.60 a month. Watching your spend never costs you spend.

On paid plans we poll usage hourly and estimate live spend against a maintained price sheet, then reconcile nightly against the official cost report. A runaway loop that starts at 2pm triggers an alert that afternoon, not on next month's invoice.

About five minutes. Create a read-only admin key at your provider, paste it in, set a budget, and connect Slack. No new SDK to install, no proxy to run, and nothing to change in your code. (Two exceptions: for Claude Enterprise, the Analytics key must be created by your claude.ai organization's primary owner — we walk you through it; and Amazon Bedrock needs a short CloudFormation stack and Cost Explorer enabled in your AWS account, which can take up to a day to populate.)

It depends on how your team uses Claude. For API spend, Anthropic doesn't offer a narrower read-only key for its Usage & Cost endpoints — only a full Admin key can read them, and creating one requires an Organization-tier Console account. PulseMeter calls only those read-only usage and cost endpoints with it. If your team uses Claude through claude.ai seats (Enterprise plans, including Claude Code), that data lives behind a separate Analytics API key that your claude.ai organization's primary owner creates — connect it under the Claude Enterprise provider. Either way we never see prompts or completions, and every key is encrypted at rest. More on our security page.

OpenAI and Anthropic with full hourly detection, Grok (xAI) with official daily costs, Claude Enterprise: per-person usage and cost for teams on claude.ai seats, including Claude Code, ChatGPT Enterprise (beta): per-person daily credits and cost for ChatGPT and Codex seats via OpenAI's workspace Admin API, and — new, in beta — Amazon Bedrock: hourly token metrics and nightly Cost Explorer reconciliation for Claude, GPT, Llama, and Amazon Nova run through AWS, via a read-only cross-account IAM role. Also new in beta: OpenRouter, with per-key daily spend in dollars from a Management key, and Cloudflare AI Gateway, which turns the gateway's own request logs into hourly spend for every provider it fronts, and Azure OpenAI: hourly token metrics from Azure Monitor and official daily cost from Cost Management via a read-only app registration. Google Gemini, Perplexity, and Microsoft Copilot are on the roadmap and shown in the connect screen.

Every person, API key, or project whose spend PulseMeter tracks — one distinct key with usage in the last 30 days, the same list you see on your Keys page. On Claude or ChatGPT Enterprise that's a person; on OpenAI, Anthropic, Grok, or OpenRouter it's an API key or project; on Amazon Bedrock it's an AWS account; on Cloudflare it's an AI Gateway; on Azure OpenAI it's a resource. Starter includes 50 monitored keys and bills $0.75 per key beyond that; Pro includes 250 and bills $0.50 per key beyond. Overage is billed on the highest count during the billing period, and you can upgrade anytime.