feat: add Zaps.ai as native LLM provider

[argosautomation]

feat: add Zaps.ai as native LLM provider

This PR adds Zaps.ai as a first-class OpenAI-compatible provider in Moltis.

What is Zaps.ai?

Zaps.ai acts as a privacy-focused API gateway that automatically redacts Personally Identifiable Information (PII) before forwarding requests to upstream LLMs (like GPT-4o, DeepSeek, etc.). This makes it ideal for enterprise and privacy-conscious deployments where data leakage is a concern.

Implementation Details

• Provider Registration: Added OpenAiCompatDef entry to crates/agents/src/providers/mod.rs with config_name: "zaps".
• Gateway UI: Added to known_providers() in crates/gateway/src/provider_setup.rs so it appears in the provider selection dropdown.
• RAG Support: Added to EMBEDDING_CANDIDATES in crates/gateway/src/server.rs to enable automatic memory embedding fallback through the Zaps gateway.
• Health Checks: Included in doctor_commands.rs PROVIDER_ENV_MAP for moltis doctor verification.
• Configuration:
  • env_key: ZAPS_API_KEY
  • default_base_url: https://zaps.ai/v1
  • models: Empty list (uses dynamic model discovery based on upstream provider availability)
  • requires_model: true (like OpenRouter)

Verification

Ran full workspace cargo check --workspace on Linux/x86_64 build environment.

• Result: BUILD_SUCCESS
• Scope: 610+ crates compiled with zero errors. All changed crates passed type checking.

Usage

Users can enable it by adding the following to moltis.toml or setting ZAPS_API_KEY:

[providers.zaps]
enabled = true
# api_key = "gk_..."  # Optional if env var set