Skip to content

Commit 919f204

Browse files
fzowlfzowl
committed
feat(embeddings): add Voyage AI embedding models, contextualized embeddings and rerank
Adds a voyage provider to the embedding catalog (voyage-4-large, voyage-4, voyage-4-lite, voyage-code-4, voyage-finance-2, voyage-law-2 and the voyage-3.x generation) with an adapter for /v1/embeddings, the embeddings_voyage tool, BYOK key support, VOYAGE_API_KEY for hosted use and per-model pricing. Adds a Voyage AI block with contextualized embeddings (list of chunk lists per document) and rerank tools, with hosted-key billing and a string detail error extractor.
1 parent 82f8c35 commit 919f204

41 files changed

Lines changed: 1205 additions & 14 deletions

File tree

Some content is hidden

Large Commits have some content hidden by default. Use the searchbox below for content that may be hidden.

‎apps/docs/components/ui/icon-mapping.ts‎

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -820,6 +820,7 @@ export const blockTypeToIconMap: Record<string, IconComponent> = {
820820
video_generator_v3: VideoIcon,
821821
vision: EyeIcon,
822822
vision_v2: EyeIcon,
823+
voyageai: EmbeddingsIcon,
823824
wait: CirclePause,
824825
wealthbox: WealthboxIcon,
825826
webflow: WebflowIcon,

‎apps/docs/content/docs/integrations/embeddings.mdx‎

Lines changed: 31 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -35,7 +35,7 @@ Knowledge bases embed separately. Each base fixes an embedding model and vector
3535

3636
## Use the block [#usage-instructions]
3737

38-
Turn text into embedding vectors for semantic search, clustering, and similarity. Supports OpenAI, OpenRouter, Google Gemini, Cohere, and Mistral embedding models, plus embedding models on a self-hosted Ollama.
38+
Turn text into embedding vectors for semantic search, clustering, and similarity. Supports OpenAI, OpenRouter, Google Gemini, Cohere, Mistral, and Voyage AI embedding models, plus embedding models on a self-hosted Ollama.
3939

4040
## Actions
4141

@@ -189,6 +189,36 @@ Generate embeddings from text using Mistral's embedding models
189189

190190
</ReferenceDetails>
191191

192+
### Voyage AI Embeddings
193+
194+
Generate embeddings from text using Voyage AI's embedding models
195+
196+
<ReferenceDetails>
197+
198+
#### Input
199+
200+
| Parameter | Type | Required | Description |
201+
| --------- | ---- | -------- | ----------- |
202+
| `input` | string | Yes | Text to embed, or an array of texts to embed in one call |
203+
| `model` | string | No | Embedding model to use |
204+
| `taskType` | string | No | What the embedding is for, when the model supports task conditioning: document, query, similarity, classification, or clustering |
205+
| `dimensions` | number | No | Output dimensions, when the model supports truncation. Defaults to native. |
206+
| `apiKey` | string | Yes | API key for the selected embedding provider |
207+
208+
#### Output
209+
210+
| Parameter | Type | Description |
211+
| --------- | ---- | ----------- |
212+
| `embeddings` | json | Generated embedding vectors, one per input, in input order |
213+
| `model` | string | Model used |
214+
| `provider` | string | Provider used |
215+
| `dimensions` | number | Dimensionality of each returned vector |
216+
| `usage` | json | Token usage |
217+
| ↳ `prompt_tokens` | number | Tokens in the input |
218+
| ↳ `total_tokens` | number | Total tokens billed |
219+
220+
</ReferenceDetails>
221+
192222
### Ollama Embeddings
193223

194224
Generate embeddings on a self-hosted Ollama server

‎apps/docs/content/docs/integrations/meta.json‎

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -296,6 +296,7 @@
296296
"uptimerobot",
297297
"vanta",
298298
"vercel",
299+
"voyageai",
299300
"wealthbox",
300301
"wealthbox-service-account",
301302
"webflow",

‎apps/docs/content/docs/integrations/voyageai.mdx‎

Lines changed: 87 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,87 @@
1+
---
2+
title: Voyage AI
3+
description: Contextualized embeddings and reranking
4+
integration:
5+
type: "voyageai"
6+
color: "#7B4DFF"
7+
---
8+
9+
import { ReferenceDetails } from "@/components/ui/reference-details"
10+
11+
12+
{/* MANUAL-CONTENT-START:intro */}
13+
[Voyage AI](https://www.voyageai.com/) provides embedding and reranking models for search and retrieval.
14+
15+
With the Voyage AI block, you can:
16+
17+
- **Embed chunks with their document context**: Send each document as a list of text chunks. Every chunk vector reflects the document around it, so retrieval finds the right passage even when the chunk alone is ambiguous.
18+
- **Rerank results**: Reorder retrieved texts by relevance to a query, so your agent reads the best matches first.
19+
20+
Create an API key in the [Voyage AI dashboard](https://dash.voyageai.com/) and paste it into the block, or store it once as a workspace key under Settings > BYOK.
21+
22+
In Sim, the Voyage AI block lets your agents index documents for search and tighten the results they retrieve before answering.
23+
{/* MANUAL-CONTENT-END */}
24+
25+
## Use the block [#usage-instructions]
26+
27+
Embed document chunks with Voyage AI contextualized models, so each chunk vector reflects the document around it, and rerank texts against a query with Voyage AI rerankers.
28+
29+
## Actions
30+
31+
### Voyage AI Contextualized Embeddings
32+
33+
Embed the chunks of one or more documents so that each chunk vector also reflects the document it came from
34+
35+
<ReferenceDetails>
36+
37+
#### Input
38+
39+
| Parameter | Type | Required | Description |
40+
| --------- | ---- | -------- | ----------- |
41+
| `input` | array | Yes | Documents to embed, as a list of lists of text chunks: one inner list per document, chunks in document order |
42+
| `model` | string | No | Embedding model: voyage-context-4 \(default\) or voyage-context-3 |
43+
| `inputType` | string | No | What the embeddings are for: document \(chunks to index\) or query \(search queries\) |
44+
| `dimensions` | number | No | Output dimensions: 256, 512, 1024 \(default\), or 2048 |
45+
| `apiKey` | string | Yes | Your Voyage AI API key |
46+
47+
#### Output
48+
49+
| Parameter | Type | Description |
50+
| --------- | ---- | ----------- |
51+
| `embeddings` | json | Embedding vectors grouped by document: one list per input document, one vector per chunk, in input order |
52+
| `model` | string | Model used |
53+
| `dimensions` | number | Dimensionality of each vector |
54+
| `totalTokens` | number | Tokens billed for the request |
55+
56+
</ReferenceDetails>
57+
58+
### Voyage AI Rerank
59+
60+
Rank a list of texts by relevance to a query with a Voyage AI reranker, most relevant first
61+
62+
<ReferenceDetails>
63+
64+
#### Input
65+
66+
| Parameter | Type | Required | Description |
67+
| --------- | ---- | -------- | ----------- |
68+
| `query` | string | Yes | The query to rank the documents against |
69+
| `documents` | array | Yes | The texts to rank, as a list of strings \(up to 1,000\) |
70+
| `model` | string | No | Reranker model: rerank-2.5 \(default\) or rerank-2.5-lite |
71+
| `topK` | number | No | Number of most relevant documents to return. Defaults to all of them. |
72+
| `apiKey` | string | Yes | Your Voyage AI API key |
73+
74+
#### Output
75+
76+
| Parameter | Type | Description |
77+
| --------- | ---- | ----------- |
78+
| `results` | array | Documents ranked by relevance, most relevant first |
79+
| ↳ `index` | number | Position of the document in the input list |
80+
| ↳ `relevanceScore` | number | Relevance of the document to the query |
81+
| ↳ `document` | string | The document text |
82+
| `model` | string | Reranker model used |
83+
| `totalTokens` | number | Tokens billed for the request |
84+
85+
</ReferenceDetails>
86+
87+

‎apps/sim/.env.example‎

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -115,6 +115,7 @@ CRON_SECRET=your_cron_secret # Use `openssl rand -hex 32` to generate. Authentic
115115
# NEXT_PUBLIC_AZURE_CONFIGURED=true # Set when Azure credentials are pre-configured above. Hides endpoint/key/version fields in Agent block UI.
116116
# COHERE_API_KEY= # Cohere API key for the Knowledge block reranker (rerank-v4.0-pro/-fast, rerank-v3.5). Alternatively set COHERE_API_KEY_1/2/3 for rotation.
117117
# NEXT_PUBLIC_COHERE_CONFIGURED=true # Set when COHERE_API_KEY (or rotation keys) are pre-configured above. Hides the Cohere API Key field on the Knowledge block UI.
118+
# VOYAGE_API_KEY= # Voyage AI API key for the Embeddings block (voyage-4-large, voyage-4, voyage-4-lite, voyage-code-4, ...) and the Voyage AI block (contextualized embeddings, rerank).
118119

119120
# Hosted tool API keys (Optional - lets Sim supply the key so users don't have to bring their own).
120121
# Each provider reads `{PREFIX}_COUNT` then `{PREFIX}_1..N`, distributing requests round-robin across the keys.

‎apps/sim/app/workspace/[workspaceId]/settings/components/byok/byok.tsx‎

Lines changed: 9 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -13,6 +13,7 @@ import {
1313
ContextDevIcon,
1414
DatagmaIcon,
1515
DropcontactIcon,
16+
EmbeddingsIcon,
1617
EnrowIcon,
1718
ExaAIIcon,
1819
FalIcon,
@@ -106,6 +107,13 @@ const PROVIDERS: (BYOKManagerProvider & { id: BYOKProviderId })[] = [
106107
description: 'LLM calls and Knowledge Base OCR',
107108
placeholder: 'Enter your API key',
108109
},
110+
{
111+
id: 'voyage',
112+
name: 'Voyage AI',
113+
icon: EmbeddingsIcon,
114+
description: 'Embeddings, contextualized embeddings and reranking',
115+
placeholder: 'pa-...',
116+
},
109117
{
110118
id: 'zai',
111119
name: 'Z.ai',
@@ -382,6 +390,7 @@ const PROVIDER_SECTIONS: BYOKProviderSection[] = [
382390
'mistral',
383391
'zai',
384392
'cohere',
393+
'voyage',
385394
'xai',
386395
'kimi',
387396
'typesafe',

‎apps/sim/blocks/blocks/embeddings.ts‎

Lines changed: 4 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -33,6 +33,7 @@ const TOOL_ID_BY_PROVIDER: Record<EmbeddingBlockProvider, string> = {
3333
gemini: 'embeddings_gemini',
3434
cohere: 'embeddings_cohere',
3535
mistral: 'embeddings_mistral',
36+
voyage: 'embeddings_voyage',
3637
ollama: 'embeddings_ollama',
3738
}
3839

@@ -42,6 +43,7 @@ const PROVIDER_LABELS: Record<EmbeddingBlockProvider, string> = {
4243
gemini: 'Google Gemini',
4344
cohere: 'Cohere',
4445
mistral: 'Mistral',
46+
voyage: 'Voyage AI',
4547
ollama: 'Ollama',
4648
}
4749

@@ -164,7 +166,7 @@ export const EmbeddingsBlock: BlockConfig = {
164166
description: 'Generate embeddings',
165167
authMode: AuthMode.ApiKey,
166168
longDescription:
167-
'Turn text into embedding vectors for semantic search, clustering, and similarity. Supports OpenAI, OpenRouter, Google Gemini, Cohere, and Mistral embedding models, plus embedding models on a self-hosted Ollama.',
169+
'Turn text into embedding vectors for semantic search, clustering, and similarity. Supports OpenAI, OpenRouter, Google Gemini, Cohere, Mistral, and Voyage AI embedding models, plus embedding models on a self-hosted Ollama.',
168170
category: 'tools',
169171
integrationType: IntegrationType.AI,
170172
docsLink: 'https://docs.sim.ai/integrations/embeddings',
@@ -237,6 +239,7 @@ export const EmbeddingsBlock: BlockConfig = {
237239
'embeddings_gemini',
238240
'embeddings_cohere',
239241
'embeddings_mistral',
242+
'embeddings_voyage',
240243
'embeddings_ollama',
241244
],
242245
config: {

0 commit comments

Comments
 (0)