| name | free-claude-code |
| description | Routes Claude-Code–style coding agents through NVIDIA's free OpenAI-compatible gateway. In OctoCode this is wired as the `nvidia-free` system fallback provider — used when the operator's configured primary provider is unreachable. |
| source | https://github.com/loonghao/free-claude-code |
| integration | system-fallback-provider |
| status | active-provider |
| registered-provider-id | nvidia-free |
free-claude-code
Promoted to active-provider in P13. The upstream project routes agent
traffic through NVIDIA's free inference gateway; OctoCode adopts the same
underlying endpoint (https://integrate.api.nvidia.com/v1) via the existing
OpenAiCompatibleProvider, registered under the provider id
nvidia-free.
Why registered
Operators routinely hit provider outages (quota exhaustion, Anthropic 529s,
local model crashes). Without a fallback, OctoCode's turn stalls and the
WebUI stream indicator sits at running until the recovery-refresh poll
fires. Adding a keyless free fallback converts that failure mode into a
degraded-but-working response.
Integration shape
- Not inserted automatically into every provider's fallback chain —
that would change long-standing test contracts and silently leak
traffic to a third-party endpoint.
- Is available as a top-level provider id (
nvidia-free). Operators
opt in either by:
- Setting
OCTOCODE_PROVIDER=nvidia-free for a given session, or
- Configuring their primary provider chain to terminate in
nvidia-free instead of stub (follow-up in P14).
Environment variables
| Variable | Purpose | Default |
|---|
OCTOCODE_NVIDIA_FREE_BASE_URL | Override the gateway endpoint | https://integrate.api.nvidia.com/v1 |
OCTOCODE_NVIDIA_FREE_API_KEY | Optional API key for higher rate limits | (none; keyless free tier) |
Safety notes
- NVIDIA's gateway is a third-party service. Never route traffic
through
nvidia-free for sessions that carry secrets, credentials, or
customer PII. Operators are responsible for redaction at the
coordinator layer.
- The provider inherits OctoCode's circuit breaker, rate limiter, and
permission gates — no new tool capability is exposed.
- Fallback is bounded: the chain is
[nvidia-free → stub], so if
NVIDIA is unreachable the stub guarantees determinism rather than a
hang.
Roadmap
- P14: allow operators to append
nvidia-free to any other
provider's chain via a single config key (e.g.
fallback_chain = ["anthropic", "nvidia-free", "stub"]).
- P14: emit an event-feed entry
provider_fallback_engaged whenever
the chain descends past the primary.
- P15: ship a health-probe that pings
/v1/models once per scrape
interval so /metrics can expose octocode_provider_up{id="nvidia-free"}.
