For builders
You build with AI and can't fully review what it writes. SID routes the
model, then checks the code before it ships.
The safety net: POST /v1/agents/run
An agent run is not just a completion. When a change is proposed, SID verifies it
before the run can claim success:
- Independent-model diff review. A second, independent model reads
the diff and the task and reports whether the change actually does the work.
- Repo tests in a sandbox. The project's own tests run inside an
isolated sandbox — never on the gateway host — and the run fails if they fail.
- Fail-closed BellerBuild security scan. The result is scanned for
the things that get people breached (exposed rows, secrets in the tree, widened
access). If the scan cannot complete, the run does not pass.
SID verifies the change. It does not promise that no bug can ever reach production;
the point is that the code is checked before it ships, not after.
One endpoint, automatic routing
Point any OpenAI-compatible tool at one base URL and one key. With
model: "auto", SID classifies the request and routes it to the best model
for the task, with an ordered fallback chain if the first candidate fails. You can also
pin a model or send your own ordered fallback list.
- Open-weight models on SID credits — one balance, metered per
request, no separate provider accounts.
- Bring your own key (BYOK) for closed models — your request goes
to the provider you connected, under your key; SID never resells those models.
- Spend caps — set a ceiling and the gateway stops at it instead
of running up a bill.
Works with the tools you already use
SID is a drop-in OpenAI-compatible backend for Cursor, Continue, Aider, Cline,
OpenCode, Zed, Claude Code and Codex. Install or connect instructions are in the
API docs.
When SID isn't for you
Be honest about the fit. SID is built around routing and verification, not around
being a marketplace of everything. Choose something else if:
- you need hundreds of raw models at wholesale token prices with no routing or
verification layer — a pure model router is the better fit;
- you need to self-host the inference yourself — SID is a hosted gateway, not a
self-hosted stack;
- you need a contractual enterprise SLA today — we publish our current posture
honestly rather than imply an SLA we have not signed.
If none of those apply, start with the free guest demo or
create a free account.