Frontier models at cost, with a spend cap that actually stops.

One endpoint for Claude, Kimi and low-cost open models on AWS Bedrock, speaking both the OpenAI and Anthropic APIs. No markup on tokens. Every request reports what it cost, and when you reach your limit, requests stop.

• At costInference is billed at Bedrock’s own per-token prices, nothing added.
• A real limitMonthly or lifetime caps, overall or per key, enforced before a request ever reaches the model.
• Cost on every responseEach response says exactly what it cost, down to the cached token.
• Both APIsOpenAI chat completions for every model, and the Anthropic Messages API for Claude — so Claude Code and the Anthropic SDKs work unchanged.
• Works with your agentsStreaming, tool calls, thinking and prompt caching, through the API your tools already speak.

# OpenAI-compatible, any model
curl https://api.glimpsed.ai/v1/chat/completions \
  -H "Authorization: Bearer $GLIMPSED_AI_KEY" \
  -d '{"model": "anthropic/claude-sonnet-5.5", "messages": [...]}'

# Anthropic Messages, Claude — or just point Claude Code at it
ANTHROPIC_BASE_URL=https://api.glimpsed.ai claude

X-Glimpsed-Cost-USD: 0.0001628