Skip to content

Releases

v2.2.0

Choose a tag to compare

@trefeon trefeon released this 17 Sep 18:49

Fix: false rejections no longer kill models

  • Upstream sensitive_words_detected 500s are prompt content-filter rejections, not outages: passed through with no retry, no model lock, no breaker penalty (IsContentFilterRejection).
  • Recovery probe carries no API key, so any 4xx (keyless 401) now clears the model lock instead of extending it forever. Previously one bad prompt hid deepseek-v4-flash from /v1/models for 10+ hours.
  • availableModels now counts healthy models, matching /v1/models.

Feat: default safe persona + gpt-6-astra

  • Unset INJECT_SYSTEM_PROMPT injects a benign engineering-framing persona into every request (custom text overrides, empty disables).
  • Added gpt-6-astra (1.05M ctx, 128K out) to the default model list; specs researched via OpenRouter.

Found while mapping: the provider flags the canonical 'You are a helpful assistant' string in user content — reword to 'coding/useful assistant'.

v2.1.0 — opencode spoof default + profile registry

Choose a tag to compare

@trefeon trefeon released this 01 Sep 13:13

Added

  • SPOOF_PROFILE env (default opencode) — profile registry in internal/auth/profile.go with 16 profiles: opencode (default, opencode/1.18.25 + interleaved-thinking — validated against reference/opencode session/llm/request.ts:18), claude-code (legacy claude-cli/2.1.92 + X-Stainless-*), codex, qwen, cline, roo, kilo, cursor, trae, pi, openclaw, hermes, droid, copilot, gemini, generic. WAF bypass verified 5 models × 2 profiles (opencode vs claude-code) — both pass for claude-opus-4-8, claude-opus-5, deepseek-v4-flash, glm-5.3, gpt-5.6-sol; generic blocked.
  • reference/opencode shallow clone (v1.18.25) for header sourcing, internal/auth/profile.go extensible via SPOOF_PROFILE
  • Config: SPOOF_PROFILE validated via auth.IsValidProfile, NormalizedSpoofProfile() → auth.NormalizeProfile, Upstream() → net.JoinHostPort
  • Handler/Server: profile-aware headers (/v1/messages Anthropic vs /v1/chat/completions OpenAI), health probe uses SpoofHeadersForProfile, fixed WaitGroup bug
  • e2e: TestSpoofHeadersInjected now opencode (opencode UA, minimal beta, no X-Stainless), added TestSpoofHeadersClaudeCode/TestGenericSpoofHeadersClaudeCode (70+ E2E pass)

Changed

  • Default spoof claude-code → opencode (lighter, whitelisted by AgentRouter per ps.air-outer.com/docs/opencode.html; claude-code remains via SPOOF_PROFILE=claude-code)
  • MODELS_CSV default 3 → 5 (deepseek-v4-flash, glm-5.3, gpt-5.6-sol added)
  • .env.example, docker-compose.yml (add SPOOF_PROFILE), README (WAF bypass, model-aware headers, architecture, config table)

Fixed

  • Docker not updated on acerblue (was 2d old a0a8cff, 3 models) → now 6c3598e, 5 models, wafCookie:true, SPOOF_PROFILE=opencode
  • go vet hostport (IPv6 bracket), e2e broken env/ lines

Binaries

  • dist/proxy-linux-amd64, proxy-linux-arm64, proxy-darwin-arm64, proxy.exe (Windows amd64, Go 1.26, static, -s -w, ~7MB)
  • internal/dashboard, logstore, checkin, counttokens, usage (previously untracked) included

Docs

  • All docs refreshed: README.md, .env.example, docker-compose.yml, docs/panduan-9router.md, docs/development.md
  • reference/opencode + reference/agentrouter-docs for per-CLI header research

Go rewrite — v2.0.0

Choose a tag to compare

@trefeon trefeon released this 14 Aug 12:44

Full rewrite in Go: fast cross-platform proxy, ~15-25MB static binary, distroless Docker image, systemd/Windows-service installers.

  • Same features as the Node version: Claude Code header spoofing, WAF cookie management, SSE streaming with keepalives + format-aware EOM, circuit breaker, auto model health, prompt injection, model discovery
  • ~222 Go tests (62 E2E + 7 issue regressions + unit suites)
  • Installers: scripts/install.sh --systemd | scripts/install.ps1 -Service (download this binary)