memory-lancedb-pro fails with Voyage embeddings (voyage-4-large / voyage-4) returning HTTP 400 during recall

[tkhs101]

Plugin Version

1.1.0-beta.9

OpenClaw Version

OpenClaw 2026.3.24 (cff6dc9)

Bug Description

安装插件后电报直接瘫痪，TUI对话还可用，配置是
Embedding: voyage-4
Rerank: rerank-2.5
LLM: gpt-5.4
检索: hybrid

下面是bot的回复
我刚查了这个插件源码，问题不是模型名本身，而是它的 embedding 请求是按 OpenAI-compatible 方式发的，固定会发这类 payload：

input
model
encoding_format: "float"
可选 dimensions
但 Voyage 的 embeddings 接口通常还要求/期望自己的参数形态，至少这个插件当前这版没有专门做 Voyage embedding 适配，所以才会持续 400。

Using memory-lancedb-pro@1.1.0-beta.9 with Voyage embeddings fails at runtime.
Plugin can register after manually setting embedding.dimensions = 1024, but actual embedding calls during recall / smart extraction fail with:

Failed to generate embedding from api.voyageai.com: 400: 400 status code (no body)

Environment

OpenClaw 2026.3.24
memory-lancedb-pro 1.1.0-beta.9
Windows 10 x64
Node 24.14.1
Config

embedding provider: openai-compatible
embedding baseURL: https://api.voyageai.com/v1
embedding model tested:
voyage-4-large
voyage-4
rerank:
provider: voyage
model: rerank-2.5
llm:
gpt-5.4
Observed behavior

Direct manual API check to Voyage embeddings endpoint returns HTTP 200
Direct manual API check to Voyage rerank endpoint returns HTTP 200
Plugin registration succeeds only after manually adding:
embedding.dimensions: 1024
But runtime logs still show:
memory-lancedb-pro: recall failed: Error: Failed to generate embedding from api.voyageai.com: 400
memory-pro: smart-extractor: failed to process candidate ...
Likely issues

Voyage embedding models are missing from built-in dimension mapping
Generic OpenAI-compatible embedding payload used by the plugin seems incompatible with Voyage embeddings API
Expected

Either Voyage embeddings should work
Or plugin should fail early with a clear validation error
Or documentation should state Voyage is supported only for reranking, not embeddings

Expected Behavior

希望能够使用voyage作为embedding《它的免费额度超高

Steps to Reproduce

先安装了对应skill
再把我想要的配置
Embedding: voyage-4
Rerank: rerank-2.5
LLM: gpt-5.4
检索: hybrid

再配上API一起发给openclaw帮我安装，然后就这样了

Error Logs / Screenshots

Embedding Provider

None

OS / Platform

No response

[AliceLJY]

Thanks for the detailed report, @tkhs101. This is a known limitation — the plugin sends embedding requests using the OpenAI-compatible payload format (input, model, encoding_format: "float"), which Voyage's API does not fully accept, resulting in the HTTP 400. PR #377 (separate internal schema dimensions from embedding API request dimensions) may be related but won't fully resolve Voyage compatibility. For now, Voyage is only reliably supported as a rerank provider. We'll look into either adding a Voyage-specific embedding adapter or documenting this limitation clearly. As a workaround, consider using a different embedding provider (e.g., OpenAI, Zhipu, or Ollama) alongside Voyage reranking.