This is an early release preview. You may encounter bugs.
Ccflare logo
Unclaimed

Tool inference monitoring

Ccflare

Rate limits and opaque failures are a primary scaling bottleneck for Claude workloads. ccflare sits in front of Anthropi

C 52/100 GitHub score ? This grade is derived from GitHub signals, not user votes. Open for the full breakdown.
No votes yet

01 / About

What Ccflare is.

ccflare is a proxy that sits in front of Anthropic and OpenAI. It forwards requests natively — provider payloads are not translated — while routing by URL prefix, load-balancing across multiple accounts, and recording full request history, rate-limit state, and usage analytics.

Requests go to /v1/{provider}/*, and the prefix is stripped exactly once before forwarding: /v1/anthropic/v1/messages reaches the Anthropic messages endpoint and /v1/openai/chat/completions reaches the OpenAI chat-completions endpoint, with the Responses API handled the same way. Clients are pointed at it by setting ANTHROPICBASEURL or OPENAIBASEURL; both can be configured at once, and account selection stays isolated per provider.

A third family of routes, /v1/ccflare/*, keeps the client-facing schema but picks a provider family from the model prefix — openai/ prefers a codex account then openai, and anthropic/ prefers claude-code then anthropic. Accounts are added as API keys through the management API or interactively over OAuth from the CLI and terminal interface, and a request is retried against another account when one is rate limited.

The management API covers the account lifecycle (create, rename, pause, resume, delete), recent and detailed request records, live request and log streams over server-sent events, aggregated analytics and usage stats, load-balancing strategy selection, data-retention settings, and cleanup and database compaction. A browser dashboard, a terminal interface, and a server-only mode share the same backend.

Features

  • Native passthrough: Anthropic requests stay Anthropic and OpenAI requests stay OpenAI, with no payload translation
  • Prefix routing: /v1/anthropic/ and /v1/openai/, with the provider prefix stripped once before forwarding
  • Compatibility routes: /v1/ccflare/* selects a provider family from a family-prefixed model id
  • Account failover: another account is retried when one provider account is rate limited
  • Flexible auth: API-key accounts through the management API, or OAuth accounts set up from the CLI
  • Built-in observability: dashboard, request history, analytics, logs, and a health endpoint
  • Live streams: request and log streams over server-sent events
  • Load-balancing strategies: listed and switched at runtime through the config API
  • Data retention: retention settings, cleanup runs, and database compaction

02 / Discussion CREDIBILITY-GATED

Discussion

Reading is open to everyone. Posting and voting need a verified identity or a GitHub grade of B or higher.

  • No discussions yet.

03 / Build

Build with Ccflare.

Browse the catalogue for frameworks, tools, and harnesses, each scored on real GitHub credibility.

Get Ccflare →

Browse the catalogue