Skip to content
BASS

Providers

Provider Auth History

Section titled “Provider Auth History”

Full provider history and lessons. Quick ref in MEMORY.md.

Current Provider Status

Section titled “Current Provider Status”

As of 2026-03-01

anthropic/

Section titled “anthropic/”
  • MAX plan OAuth token (ANTHROPIC_MAX_TOKEN). Primary for ALL agents.
  • Hard rules: memory/ops/anthropic-auth-rules.md
  • ANTHROPIC_FRANKEL_KEY — Paid API key ($25 credit). Frankel-ONLY. NEVER use for general agent work.
  • ANTHROPIC_ADMIN_API_KEY — Admin API for usage/cost reporting only.

openai-codex/

Section titled “openai-codex/”
  • OAuth (ChatGPT Plus). ONLY working OpenAI provider.
  • Models: gpt-5.3-codex, gpt-5.3-codex-spark, gpt-5.1-codex-mini
  • Config: ~/.codex/config.toml, multi_agent=true

openai/

Section titled “openai/”
  • API key set but NO QUOTA (exceeded). Do NOT use as fallback.
  • Auth:yes ≠ quota. This provider will fail silently.

openrouter/

Section titled “openrouter/”
  • API key, works but limited credits. Use sparingly.

google/

Section titled “google/”
  • AI Studio API key (GEMINI_API_KEY env). FREE TIER — quota exhausts quickly on Pro models.
  • google/gemini-3.1-pro-preview — exists upstream but unstable/unsupported in OpenClaw runs.
  • google/gemini-3-pro-preview — free-tier quota exhausted.
  • Use for light work only. Prefer google-antigravity/ for sustained agent workloads.

google-antigravity/

Section titled “google-antigravity/”
  • OAuth (2 accounts). Works. Gemini + Claude models.
  • google-antigravity/gemini-3.1-pro-preview — stable, higher rate limits than google/
  • google-antigravity/gemini-3-pro-high — returns deprecation message. Don’t use.
  • google-antigravity/gemini-3-flash — current stable Scout primary

google-gemini-cli/

Section titled “google-gemini-cli/”
  • OAuth (wired from CLI creds, 2026-02-21). Works. Gemini models.

minimax-portal/

Section titled “minimax-portal/”
  • OAuth. Free tier. Works.
  • 88% hallucination rate. Heartbeats only.

Incident History

Section titled “Incident History”

2026-02-27: anthropic-direct Leak (~$15 lost)

Section titled “2026-02-27: anthropic-direct Leak (~$15 lost)”
  • anthropic-direct provider added at some point referencing a direct API key.
  • Leaked ~$15 in API credits in 2 days before discovery.
  • Removed. Never add again.
  • Lesson: Always verify auth profiles reference existing env vars before gateway restart. Unknown env var = crash-loop risk OR silent credential leak.

2026-02-21: Scout Routing Investigation

Section titled “2026-02-21: Scout Routing Investigation”
  • Tried google/gemini-3.1-pro-preview — exists upstream but unstable in OpenClaw runs.
  • Tried google/gemini-3-pro-preview — free-tier quota exhausted.
  • Tried google-antigravity/gemini-3-pro-high — returns deprecation message.
  • Resolved: Scout primary = google-antigravity/gemini-3-flash with codex/sonnet fallbacks.

2026-02-20: Agent-to-Agent Comms Enabled

Section titled “2026-02-20: Agent-to-Agent Comms Enabled”
  • tools.agentToAgent = true, tools.sessions.visibility = all
  • Cross-agent sessions_send working from this date.
  • Discovered nested lane concurrency=1 issue (hardcoded, no config knob).
  • GitHub issue: openclaw/openclaw#22167

2026-02-20: Sudo NOPASSWD Configured

Section titled “2026-02-20: Sudo NOPASSWD Configured”
  • /etc/sudoers.d/winnie-packages — limited package management commands only.

2026-02-18: Model Catalog Established

Section titled “2026-02-18: Model Catalog Established”
  • anthropic/claude-sonnet-4-6 added.
  • Per-agent models set per Chris’s spec.

Rule: Auth ≠ Quota

Section titled “Rule: Auth ≠ Quota”

Before assigning any provider as primary or fallback:

  1. Test a real completion (not just auth ping)
  2. Verify quota is available
  3. Test under load if it’s for a critical path

Providers that look authenticated but have no quota: openai/ (exceeded), google/ (exhausts fast on Pro).

See Also

Section titled “See Also”