Add custom provider

[madebyaris]

Summary by cubic

Adds a Custom provider so you can connect to any OpenAI- or Anthropic-compatible endpoint. Includes a TUI wizard, /custom command, config/env support, key validation, and docs.

• New Features

  • New provider: Custom (OpenAI- or Anthropic-compatible) with model/context limits and model catalog support.
  • CLI: /custom <openai|anthropic> <base-url> [api-key] [model]; /provider custom now supported.
  • TUI: Provider picker shows “Custom” and “Add custom provider…” wizard; palette entry added; contextual help with keybindings.
  • Config: new [provider.custom] block (compatibility, base_url, model, temperature, api_key_env); env vars CUSTOM_PROVIDER_API_KEY, CUSTOM_PROVIDER_BASE_URL, CUSTOM_PROVIDER_MODEL, CUSTOM_PROVIDER_COMPATIBILITY.
  • Validation: API key checks for Custom endpoints (OpenAI or Anthropic wire protocol).
  • Docs/README updated; tests added/updated across CLI, common, core, and runtime.

• Migration

  • TUI: run /provider → “Add custom provider…” and follow the steps.
  • CLI: run /custom openai https://your-endpoint key model (or anthropic).
  • Config (optional):
    • [provider] default = "custom"
    • [provider.custom] compatibility = "openai" | "anthropic"; base_url = "..."; model = "...".
    • Or set env vars: CUSTOM_PROVIDER_API_KEY, CUSTOM_PROVIDER_BASE_URL, CUSTOM_PROVIDER_MODEL, CUSTOM_PROVIDER_COMPATIBILITY.

^{Written for commit 97cc35e. Summary will update on new commits.}

[cubic-dev-ai]

2 issues found across 19 files

Prompt for AI agents (unresolved issues)

Check if these issues are valid — if so, understand the root cause of each and fix them. If appropriate, use sub-agents to investigate and fix each issue separately.


<file name="docs/plans/custom-provider-compatibility.md">

<violation number="1" location="docs/plans/custom-provider-compatibility.md:5">
P2: This plan describes `Custom` provider support as a new feature even though the repo docs already show it exists; reframe it as extending/refining existing behavior to avoid duplicate implementation work.</violation>
</file>

<file name="crates/common/src/model_caps.rs">

<violation number="1" location="crates/common/src/model_caps.rs:45">
P2: `ProviderKind::Custom` misses the `qwen-vl` multimodal heuristic, causing image-capable Qwen-VL models to be misclassified as non-vision.</violation>
</file>

_{Reply with feedback, questions, or to request a fix. Tag @cubic-dev-ai to re-run a review.}

[cubic-dev-ai]

P2: This plan describes Custom provider support as a new feature even though the repo docs already show it exists; reframe it as extending/refining existing behavior to avoid duplicate implementation work.

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. At docs/plans/custom-provider-compatibility.md, line 5:

<comment>This plan describes `Custom` provider support as a new feature even though the repo docs already show it exists; reframe it as extending/refining existing behavior to avoid duplicate implementation work.</comment>

<file context>
@@ -0,0 +1,21 @@
+
+## Goal
+
+Add a configurable `Custom` provider so users can route `nca` to arbitrary endpoints and choose whether the endpoint speaks an OpenAI-compatible or Anthropic-compatible API.
+
+## Implementation
</file context>

Suggested change

	Add a configurable `Custom` provider so users can route `nca` to arbitrary endpoints and choose whether the endpoint speaks an OpenAI-compatible or Anthropic-compatible API.
	Extend the existing `Custom` provider so users can route `nca` to arbitrary endpoints with explicit OpenAI-compatible or Anthropic-compatible behavior.

[cubic-dev-ai]

P2: ProviderKind::Custom misses the qwen-vl multimodal heuristic, causing image-capable Qwen-VL models to be misclassified as non-vision.

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. At crates/common/src/model_caps.rs, line 45:

<comment>`ProviderKind::Custom` misses the `qwen-vl` multimodal heuristic, causing image-capable Qwen-VL models to be misclassified as non-vision.</comment>

<file context>
@@ -35,6 +35,15 @@ pub fn model_accepts_native_images(kind: ProviderKind, model: &str) -> bool {
+                || m.contains("claude-3")
+                || m.contains("claude-4")
+                || m.contains("gemini")
+                || m.contains("vision")
+        }
     }
</file context>

Suggested change

	|| m.contains("vision")
	|| m.contains("qwen-vl")
	|| m.contains("vision")