Basevex AI documentation
Basevex hosts LoRA adapters on shared base models and serves them through an OpenAI-compatible API. You upload an adapter, create a deployment with an alias, and call it with any OpenAI SDK by setting model to that alias.
Concepts
| Organization | Your team’s workspace. Everything below belongs to one organization. Members are owners or members. |
|---|---|
| Base model | A model we serve, pinned to an exact revision. Your adapter must be trained on it. |
| Adapter | Your LoRA weights (adapter_config.json + adapter_model.safetensors), validated before use. |
| Deployment | An alias (e.g. contract-review) pointing at a validated adapter, with token and rate limits. The alias is the model you send. |
| Client | One of your end customers. Assign adapters, deployments and API keys to a client to attribute usage and cost. |
| API key | A secret starting with bvx_, shown once at creation. Can be limited to a client or specific deployments. |
| Sandbox | Organizations without a paid plan. The API works, but responses are synthetic, labeled and never billed. |
Start here
QuickstartFrom sign-up to your first request with the OpenAI SDK.Preparing adaptersWhich files to upload and how they are validated.Inference APIRequest and response format, streaming, supported parameters.Clients & attributionAttribute usage to your clients and export reports.Usage & billingPlans, token metering, the sandbox and spend caps.Error codesEvery error code the API returns and what to do about it.