everruns-provider

Lean provider and LLM abstractions shared by the Everruns Framework and provider crates.

everruns-provider owns credential-free model identity, the ChatDriver boundary, provider assembly, protocol drivers, stream/retry helpers, typed IDs, credential schemas, and the LLM error taxonomy. Model profile types and data (ModelProfile, ServiceKind, and the built-in profile registry) live in everruns-model-profiles and are re-exported here for source compatibility.

It is a focused implementation crate in the Everruns ecosystem. Framework application authors normally use the curated surface through everruns; provider implementers and low-level hosts depend here directly to avoid pulling in the agent-loop kernel.

The default http feature provides the shared protocol implementations and installs the Rustls crypto provider before constructing their clients. Contract- only consumers can use default-features = false to avoid HTTP and TLS dependencies entirely. The tls-aws-lc-rs feature exposes the idempotent startup initializer independently for binaries that assemble multiple TLS stacks.

Quick Example

use everruns_provider::ModelSpec;

let model = ModelSpec::on("company-gateway", "assistant-v2");
assert_eq!(model.provider.as_str(), "company-gateway");
assert_eq!(model.model, "assistant-v2");

What It Provides

• Open ChatDriver and provider registry contracts
• Credential-free ModelSpec and redacting endpoint/auth values
• Shared OpenAI and Open Responses protocol implementations
• Streaming, retry, model discovery/profile, and error helpers
• Provider-oriented tool and credential schema values

Documentation

• Framework models and providers
• Custom providers
• API reference

License

Licensed under the MIT License.