feat(providers): add Liquid AI (LFM2) as OpenAI-compatible provider

[qhkm]

Summary

• Adds Liquid AI as a first-class OpenAI-compatible provider (base URL https://labs.liquid.ai/api/v1) — aligned with ZeptoClaw's edge/IoT thesis and the LFM2 model family (LFM2-1.2B-Tool, LFM2.5-1.2B-Instruct, LFM2-24B-A2B, etc.)
• Pure registry plumbing — no new provider code, no new dependencies, reuses existing OpenAI-compat infrastructure exactly like novita / kimi / qianfan
• Model keywords (liquid, lfm, lfm2) allow auto-inference from model IDs like lfm2-24b-a2b

Related Issue

Closes #541

Files changed (5 locations, 4 files)

1. src/config/types.rs — added pub liquid: Option<ProviderConfig> to ProvidersConfig with #[serde(default)]
2. src/providers/registry.rs — added ProviderSpec entry to PROVIDER_REGISTRY + match arm in provider_config_by_name()
3. src/config/mod.rs — added env handlers for ZEPTOCLAW_PROVIDERS_LIQUID_{API_KEY,API_BASE,MODEL} (with LIQUID_API_KEY fallback)
4. src/providers/mod.rs — added "liquid" to RUNTIME_SUPPORTED_PROVIDERS (required by the existing test_runtime_supported_constant_stays_in_sync sync test — this was the only deviation from the 5-location spec)
5. Tests: test_liquid_resolves_with_default_base_url, test_liquid_model_inference (in registry.rs), and test_liquid_env_override_api_key (in config/mod.rs)

Pre-submit Checklist

• I branched from upstream/main
• This PR contains only commits related to this change
• cargo fmt --check passes
• cargo clippy -- -D warnings passes
• cargo nextest run --lib passes (3420 passed, 6 skipped)
• cargo test --doc passes (128 passed)
• I added or updated tests for my changes (3 new tests)
• New constants/limits are shared (not duplicated across files)
• No new dependencies unless necessary (we target ~6 MB binary)

Security Considerations

N/A — adds a config-only provider entry; the Liquid AI API key is handled by the same env/config path as all existing providers, with the existing redaction and env-override precedence rules.

Test Plan

Automated:

• cargo nextest run --lib — full suite green (3420/3420)
• cargo test --doc — 128 passed
• New tests cover: default base URL resolution, model-keyword inference for lfm/lfm2/liquid/liquid/lfm2-*, and ZEPTOCLAW_PROVIDERS_LIQUID_API_KEY env override

Smoke test (requires a Liquid Labs API key from https://labs.liquid.ai):

export LIQUID_API_KEY=<key>
cargo run --release -- agent --provider liquid --model lfm2-24b-a2b -m "ping"
# optional: test the tool-specialized 1.2B edge variant
cargo run --release -- agent --provider liquid --model lfm2-1.2b-tool -m "What is 2+2?"

Expected: normal chat completion response; tool calling works for lfm2-*-tool and lfm2-24b-a2b (native function calling per Liquid AI docs).

Summary by CodeRabbit

• New Features

  • Liquid AI provider is now available for use in the application
  • Configuration support for API key, base URL, and model selection
  • Environment variables can override Liquid provider settings

• Tests

  • Added verification tests for Liquid provider configuration and model routing

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro

Run ID: dbb81a2b-db83-421f-bc5d-858732aa7079

📥 Commits

Reviewing files that changed from the base of the PR and between 1aa1a4e and 4739f8c.

📒 Files selected for processing (4)

• src/config/mod.rs
• src/config/types.rs
• src/providers/mod.rs
• src/providers/registry.rs

---

📝 Walkthrough

Walkthrough

The pull request adds Liquid AI (LFM) as a runtime-supported provider by extending configuration infrastructure with environment variable overrides, creating a new provider specification entry in the registry with OpenAI-compatible API routing, and registering it in the supported providers constant.

Changes

Cohort / File(s)	Summary
Configuration Infrastructure src/config/mod.rs, src/config/types.rs	Added environment variable override logic for Liquid API key (with fallback), API base, and per-provider model selection; introduced optional liquid field to ProvidersConfig; includes unit test validating env override application.
Provider Registration src/providers/mod.rs, src/providers/registry.rs	Registered "liquid" in RUNTIME_SUPPORTED_PROVIDERS constant and added new ProviderSpec entry with model keyword mapping (lfm, lfm2 families), OpenAI-compatible backend routing, default base URL (https://labs.liquid.ai/api/v1), and API-key requirement; includes unit tests for default base URL resolution and model inference routing.

Estimated code review effort

🎯 2 (Simple) | ⏱️ ~10 minutes

Possibly related PRs

• feat(providers): add Google Vertex AI provider #447: Makes parallel code-level additions to the same configuration and provider registration surfaces (env overrides, ProvidersConfig, RUNTIME_SUPPORTED_PROVIDERS, and provider registry) to add a new runtime provider.
• feat(providers): add Novita AI provider #390: Adds a new OpenAI-compatible provider by updating identical config/env-override logic, ProvidersConfig field, and provider registry entries.
• feat(providers): add Azure OpenAI and Amazon Bedrock provider presets #211: Modifies provider wiring across config types, env-override logic, RUNTIME_SUPPORTED_PROVIDERS, and provider registry to add new provider presets.

Poem

🐰 Through liquid labs the models flow,
Edge-native LFMs all aglow,
A config hop, a test that squeaks,
The smallest models, strongest techniques!
✨ Hop-hop for Liquid's edge-native dreams!

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title accurately summarizes the main change: adding Liquid AI as an OpenAI-compatible provider with LFM2 model support.
Linked Issues check	✅ Passed	The PR fulfills Phase 1 objectives [#541]: cloud provider configured, model inference working, tool-calling via OpenAI compatibility. Phases 2 and 3 remain as future work, matching issue scope.
Out of Scope Changes check	✅ Passed	All changes align with Phase 1 scope: config/registry additions for Liquid provider, env variable handling, and unit tests. No unrelated refactoring or Phase 2/3 work present.
Docstring Coverage	✅ Passed	Docstring coverage is 88.89% which is sufficient. The required threshold is 80.00%.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch feat/liquid-provider

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}