@mastra/rag 1.2.3-alpha.0 → 1.2.3-alpha.1

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (67) hide show
  1. package/CHANGELOG.md +9 -0
  2. package/package.json +18 -5
  3. package/.turbo/turbo-build.log +0 -4
  4. package/docker-compose.yaml +0 -22
  5. package/eslint.config.js +0 -6
  6. package/src/document/document.test.ts +0 -2975
  7. package/src/document/document.ts +0 -335
  8. package/src/document/extractors/base.ts +0 -30
  9. package/src/document/extractors/index.ts +0 -5
  10. package/src/document/extractors/keywords.test.ts +0 -125
  11. package/src/document/extractors/keywords.ts +0 -126
  12. package/src/document/extractors/questions.test.ts +0 -120
  13. package/src/document/extractors/questions.ts +0 -111
  14. package/src/document/extractors/summary.test.ts +0 -107
  15. package/src/document/extractors/summary.ts +0 -122
  16. package/src/document/extractors/title.test.ts +0 -121
  17. package/src/document/extractors/title.ts +0 -185
  18. package/src/document/extractors/types.ts +0 -40
  19. package/src/document/index.ts +0 -2
  20. package/src/document/prompts/base.ts +0 -77
  21. package/src/document/prompts/format.ts +0 -9
  22. package/src/document/prompts/index.ts +0 -15
  23. package/src/document/prompts/prompt.ts +0 -60
  24. package/src/document/prompts/types.ts +0 -29
  25. package/src/document/schema/index.ts +0 -3
  26. package/src/document/schema/node.ts +0 -187
  27. package/src/document/schema/types.ts +0 -40
  28. package/src/document/transformers/character.ts +0 -267
  29. package/src/document/transformers/html.ts +0 -346
  30. package/src/document/transformers/json.ts +0 -536
  31. package/src/document/transformers/latex.ts +0 -11
  32. package/src/document/transformers/markdown.ts +0 -239
  33. package/src/document/transformers/semantic-markdown.ts +0 -227
  34. package/src/document/transformers/sentence.ts +0 -314
  35. package/src/document/transformers/text.ts +0 -158
  36. package/src/document/transformers/token.ts +0 -137
  37. package/src/document/transformers/transformer.ts +0 -5
  38. package/src/document/types.ts +0 -145
  39. package/src/document/validation.ts +0 -158
  40. package/src/graph-rag/index.test.ts +0 -235
  41. package/src/graph-rag/index.ts +0 -306
  42. package/src/index.ts +0 -8
  43. package/src/rerank/index.test.ts +0 -150
  44. package/src/rerank/index.ts +0 -198
  45. package/src/rerank/relevance/cohere/index.ts +0 -56
  46. package/src/rerank/relevance/index.ts +0 -3
  47. package/src/rerank/relevance/mastra-agent/index.ts +0 -32
  48. package/src/rerank/relevance/zeroentropy/index.ts +0 -26
  49. package/src/tools/README.md +0 -153
  50. package/src/tools/document-chunker.ts +0 -34
  51. package/src/tools/graph-rag.test.ts +0 -115
  52. package/src/tools/graph-rag.ts +0 -157
  53. package/src/tools/index.ts +0 -3
  54. package/src/tools/types.ts +0 -126
  55. package/src/tools/vector-query-database-config.test.ts +0 -190
  56. package/src/tools/vector-query.test.ts +0 -477
  57. package/src/tools/vector-query.ts +0 -171
  58. package/src/utils/convert-sources.ts +0 -43
  59. package/src/utils/default-settings.ts +0 -38
  60. package/src/utils/index.ts +0 -3
  61. package/src/utils/tool-schemas.ts +0 -38
  62. package/src/utils/vector-prompts.ts +0 -832
  63. package/src/utils/vector-search.ts +0 -130
  64. package/tsconfig.build.json +0 -9
  65. package/tsconfig.json +0 -5
  66. package/tsup.config.ts +0 -17
  67. package/vitest.config.ts +0 -8
@@ -1,145 +0,0 @@
1
- import type { TiktokenEncoding, TiktokenModel } from 'js-tiktoken';
2
- import type {
3
- TitleExtractorsArgs,
4
- SummaryExtractArgs,
5
- QuestionAnswerExtractArgs,
6
- KeywordExtractArgs,
7
- } from './extractors';
8
-
9
- export enum Language {
10
- CPP = 'cpp',
11
- GO = 'go',
12
- JAVA = 'java',
13
- KOTLIN = 'kotlin',
14
- JS = 'js',
15
- TS = 'ts',
16
- PHP = 'php',
17
- PROTO = 'proto',
18
- PYTHON = 'python',
19
- RST = 'rst',
20
- RUBY = 'ruby',
21
- RUST = 'rust',
22
- SCALA = 'scala',
23
- SWIFT = 'swift',
24
- MARKDOWN = 'markdown',
25
- LATEX = 'latex',
26
- HTML = 'html',
27
- SOL = 'sol',
28
- CSHARP = 'csharp',
29
- COBOL = 'cobol',
30
- C = 'c',
31
- LUA = 'lua',
32
- PERL = 'perl',
33
- HASKELL = 'haskell',
34
- ELIXIR = 'elixir',
35
- POWERSHELL = 'powershell',
36
- }
37
-
38
- export type ExtractParams = {
39
- title?: TitleExtractorsArgs | boolean;
40
- summary?: SummaryExtractArgs | boolean;
41
- questions?: QuestionAnswerExtractArgs | boolean;
42
- keywords?: KeywordExtractArgs | boolean;
43
- };
44
-
45
- export type BaseChunkOptions = {
46
- /**
47
- * @deprecated Use `maxSize` instead. Will be removed in next major version.
48
- */
49
- size?: number;
50
- maxSize?: number;
51
- overlap?: number;
52
- lengthFunction?: (text: string) => number;
53
- keepSeparator?: boolean | 'start' | 'end';
54
- addStartIndex?: boolean;
55
- stripWhitespace?: boolean;
56
- };
57
-
58
- export type CharacterChunkOptions = BaseChunkOptions & {
59
- separator?: string;
60
- isSeparatorRegex?: boolean;
61
- };
62
-
63
- export type RecursiveChunkOptions = BaseChunkOptions & {
64
- separators?: string[];
65
- isSeparatorRegex?: boolean;
66
- language?: Language;
67
- };
68
-
69
- export type TokenChunkOptions = BaseChunkOptions & {
70
- encodingName?: TiktokenEncoding;
71
- modelName?: TiktokenModel;
72
- allowedSpecial?: Set<string> | 'all';
73
- disallowedSpecial?: Set<string> | 'all';
74
- };
75
-
76
- export type MarkdownChunkOptions = BaseChunkOptions & {
77
- headers?: [string, string][];
78
- returnEachLine?: boolean;
79
- stripHeaders?: boolean;
80
- };
81
-
82
- export type SemanticMarkdownChunkOptions = BaseChunkOptions & {
83
- joinThreshold?: number;
84
- encodingName?: TiktokenEncoding;
85
- modelName?: TiktokenModel;
86
- allowedSpecial?: Set<string> | 'all';
87
- disallowedSpecial?: Set<string> | 'all';
88
- };
89
-
90
- export type HTMLChunkOptions = BaseChunkOptions &
91
- (
92
- | { headers: [string, string][]; sections?: never; returnEachLine?: boolean }
93
- | { sections: [string, string][]; headers?: never }
94
- ) & { returnEachLine?: boolean };
95
-
96
- export type JsonChunkOptions = BaseChunkOptions & {
97
- minSize?: number;
98
- ensureAscii?: boolean;
99
- convertLists?: boolean;
100
- };
101
-
102
- export type LatexChunkOptions = BaseChunkOptions & {};
103
-
104
- export type SentenceChunkOptions = BaseChunkOptions & {
105
- maxSize: number; // Override to make required for sentence strategy
106
- minSize?: number;
107
- targetSize?: number;
108
- sentenceEnders?: string[];
109
- fallbackToWords?: boolean;
110
- fallbackToCharacters?: boolean;
111
- };
112
-
113
- export type StrategyOptions = {
114
- recursive: RecursiveChunkOptions;
115
- character: CharacterChunkOptions;
116
- token: TokenChunkOptions;
117
- markdown: MarkdownChunkOptions;
118
- html: HTMLChunkOptions;
119
- json: JsonChunkOptions;
120
- latex: LatexChunkOptions;
121
- sentence: SentenceChunkOptions;
122
- 'semantic-markdown': SemanticMarkdownChunkOptions;
123
- };
124
-
125
- export type ChunkStrategy =
126
- | 'recursive'
127
- | 'character'
128
- | 'token'
129
- | 'markdown'
130
- | 'html'
131
- | 'json'
132
- | 'latex'
133
- | 'sentence'
134
- | 'semantic-markdown';
135
-
136
- export type ChunkParams =
137
- | ({ strategy?: 'character' } & CharacterChunkOptions & { extract?: ExtractParams })
138
- | ({ strategy: 'recursive' } & RecursiveChunkOptions & { extract?: ExtractParams })
139
- | ({ strategy: 'token' } & TokenChunkOptions & { extract?: ExtractParams })
140
- | ({ strategy: 'markdown' } & MarkdownChunkOptions & { extract?: ExtractParams })
141
- | ({ strategy: 'html' } & HTMLChunkOptions & { extract?: ExtractParams })
142
- | ({ strategy: 'json' } & JsonChunkOptions & { extract?: ExtractParams })
143
- | ({ strategy: 'latex' } & LatexChunkOptions & { extract?: ExtractParams })
144
- | ({ strategy: 'sentence' } & SentenceChunkOptions & { extract?: ExtractParams })
145
- | ({ strategy: 'semantic-markdown' } & SemanticMarkdownChunkOptions & { extract?: ExtractParams });
@@ -1,158 +0,0 @@
1
- import { z } from 'zod';
2
- import type { ChunkStrategy } from './types';
3
-
4
- function handleDeprecatedSize<T extends { size?: number; maxSize?: number }>(data: T): Omit<T, 'size'> {
5
- if (data.size !== undefined) {
6
- console.warn(
7
- '[DEPRECATION] `size` is deprecated. Use `maxSize` instead. This will be removed in the next major version.',
8
- );
9
-
10
- if (data.maxSize === undefined) {
11
- data.maxSize = data.size;
12
- }
13
- }
14
-
15
- const { size, ...rest } = data;
16
- return rest;
17
- }
18
-
19
- // Base options that apply to all strategies
20
- const baseChunkOptionsSchema = z.object({
21
- size: z.number().positive().optional(),
22
- maxSize: z.number().positive().optional(),
23
- overlap: z.number().min(0).optional(),
24
- lengthFunction: z.function().optional(),
25
- keepSeparator: z.union([z.boolean(), z.literal('start'), z.literal('end')]).optional(),
26
- addStartIndex: z.boolean().optional(),
27
- stripWhitespace: z.boolean().optional(),
28
- });
29
-
30
- // Strategy-specific schemas
31
- const characterChunkOptionsSchema = baseChunkOptionsSchema
32
- .extend({
33
- separator: z.string().optional(),
34
- isSeparatorRegex: z.boolean().optional(),
35
- })
36
- .strict();
37
-
38
- const recursiveChunkOptionsSchema = baseChunkOptionsSchema
39
- .extend({
40
- separators: z.array(z.string()).optional(),
41
- isSeparatorRegex: z.boolean().optional(),
42
- language: z.string().optional(),
43
- })
44
- .strict();
45
-
46
- const sentenceChunkOptionsSchema = baseChunkOptionsSchema
47
- .extend({
48
- maxSize: z.number().positive(),
49
- minSize: z.number().positive().optional(),
50
- targetSize: z.number().positive().optional(),
51
- sentenceEnders: z.array(z.string()).optional(),
52
- fallbackToWords: z.boolean().optional(),
53
- fallbackToCharacters: z.boolean().optional(),
54
- })
55
- .strict();
56
-
57
- // Predicate to check for Set-like objects
58
- const isSetLike = (value: unknown): value is Set<any> => {
59
- return (
60
- typeof value === 'object' &&
61
- value !== null &&
62
- typeof (value as Set<any>).has === 'function' &&
63
- typeof (value as Set<any>).add === 'function' &&
64
- typeof (value as Set<any>).delete === 'function' &&
65
- typeof (value as Set<any>).clear === 'function' &&
66
- typeof (value as Set<any>).size === 'number'
67
- );
68
- };
69
-
70
- // Zod schema for a Set or the literal 'all'
71
- const setOrAllSchema = z
72
- .any()
73
- .refine(value => value === 'all' || isSetLike(value), {
74
- message: "Must be a Set object or the literal 'all'",
75
- })
76
- .optional();
77
-
78
- const tokenChunkOptionsSchema = baseChunkOptionsSchema
79
- .extend({
80
- encodingName: z.string().optional(),
81
- modelName: z.string().optional(),
82
- allowedSpecial: setOrAllSchema,
83
- disallowedSpecial: setOrAllSchema,
84
- })
85
- .strict();
86
-
87
- const jsonChunkOptionsSchema = baseChunkOptionsSchema
88
- .extend({
89
- minSize: z.number().positive().optional(),
90
- ensureAscii: z.boolean().optional(),
91
- convertLists: z.boolean().optional(),
92
- })
93
- .strict();
94
-
95
- const htmlChunkOptionsSchema = baseChunkOptionsSchema
96
- .extend({
97
- headers: z.array(z.tuple([z.string(), z.string()])).optional(),
98
- sections: z.array(z.tuple([z.string(), z.string()])).optional(),
99
- returnEachLine: z.boolean().optional(),
100
- })
101
- .strict();
102
-
103
- const markdownChunkOptionsSchema = baseChunkOptionsSchema
104
- .extend({
105
- headers: z.array(z.tuple([z.string(), z.string()])).optional(),
106
- returnEachLine: z.boolean().optional(),
107
- stripHeaders: z.boolean().optional(),
108
- })
109
- .strict();
110
-
111
- const semanticMarkdownChunkOptionsSchema = baseChunkOptionsSchema
112
- .extend({
113
- joinThreshold: z.number().positive().optional(),
114
- encodingName: z.string().optional(),
115
- modelName: z.string().optional(),
116
- allowedSpecial: setOrAllSchema,
117
- disallowedSpecial: setOrAllSchema,
118
- })
119
- .strict();
120
-
121
- const latexChunkOptionsSchema = baseChunkOptionsSchema.strict();
122
-
123
- // Strategy-specific validation schemas
124
- const validationSchemas = {
125
- character: characterChunkOptionsSchema.transform(handleDeprecatedSize),
126
- recursive: recursiveChunkOptionsSchema.transform(handleDeprecatedSize),
127
- sentence: sentenceChunkOptionsSchema.transform(handleDeprecatedSize),
128
- token: tokenChunkOptionsSchema.transform(handleDeprecatedSize),
129
- json: jsonChunkOptionsSchema.transform(handleDeprecatedSize),
130
- html: htmlChunkOptionsSchema.transform(handleDeprecatedSize),
131
- markdown: markdownChunkOptionsSchema.transform(handleDeprecatedSize),
132
- 'semantic-markdown': semanticMarkdownChunkOptionsSchema.transform(handleDeprecatedSize),
133
- latex: latexChunkOptionsSchema.transform(handleDeprecatedSize),
134
- } as const;
135
-
136
- export function validateChunkParams(strategy: ChunkStrategy, params: any): void {
137
- const schema = validationSchemas[strategy];
138
- if (!schema) {
139
- throw new Error(`Unknown chunking strategy: ${strategy}`);
140
- }
141
-
142
- const result = schema.safeParse(params);
143
- if (!result.success) {
144
- // Extract unrecognized keys for cleaner error message
145
- const unrecognizedError = result.error.errors.find((e: any) => e.code === 'unrecognized_keys');
146
- if (unrecognizedError && 'keys' in unrecognizedError) {
147
- const keys = (unrecognizedError as any).keys.join(', ');
148
- throw new Error(`Invalid parameters for ${strategy} strategy: '${keys}' not supported`);
149
- }
150
-
151
- // Fallback to general error message for other validation issues
152
- const errorMessage = result.error.errors
153
- .map((e: any) => `${e.path.length > 0 ? e.path.join('.') : 'parameter'}: ${e.message}`)
154
- .join(', ');
155
-
156
- throw new Error(`Invalid parameters for ${strategy} strategy: ${errorMessage}`);
157
- }
158
- }
@@ -1,235 +0,0 @@
1
- import { describe, it, expect, vi, beforeEach } from 'vitest';
2
-
3
- import type { GraphChunk, GraphEdge, GraphEmbedding, GraphNode } from './';
4
- import { GraphRAG } from './';
5
-
6
- describe('GraphRAG', () => {
7
- beforeEach(() => {
8
- vi.clearAllMocks(); // Clear any mock state before each test
9
- });
10
-
11
- describe('addNode', () => {
12
- it('should throw an error if node does not have an embedding', () => {
13
- const graph = new GraphRAG();
14
- const node = {
15
- id: '1',
16
- content: 'Node 1',
17
- };
18
- expect(() => graph.addNode(node)).toThrow('Node must have an embedding');
19
- });
20
-
21
- it('should throw an error if node embedding dimension is not equal to the graph dimension', () => {
22
- const graph = new GraphRAG(2);
23
- const node: GraphNode = {
24
- id: '1',
25
- content: 'Node 1',
26
- embedding: [1, 2, 3],
27
- };
28
- expect(() => graph.addNode(node)).toThrow('Embedding dimension must be 2');
29
- });
30
-
31
- it('should add a node to the graph', () => {
32
- const graph = new GraphRAG(3);
33
- const node = {
34
- id: '1',
35
- content: 'Node 1',
36
- embedding: [1, 2, 3],
37
- };
38
- graph.addNode(node);
39
- expect(graph['nodes'].size).toBe(1);
40
- });
41
- });
42
-
43
- describe('addEdge', () => {
44
- it('should throw an error if either source or target node does not exist', () => {
45
- const graph = new GraphRAG();
46
- const edge: GraphEdge = {
47
- source: '1',
48
- target: '2',
49
- weight: 0.5,
50
- type: 'semantic',
51
- };
52
- expect(() => graph.addEdge(edge)).toThrow('Both source and target nodes must exist');
53
- });
54
-
55
- it('should add an edge between two nodes', () => {
56
- const graph = new GraphRAG(3);
57
- const node1: GraphNode = {
58
- id: '1',
59
- content: 'Node 1',
60
- embedding: [1, 2, 3],
61
- };
62
- const node2: GraphNode = {
63
- id: '2',
64
- content: 'Node 2',
65
- embedding: [4, 5, 6],
66
- };
67
- graph.addNode(node1);
68
- graph.addNode(node2);
69
- const edge: GraphEdge = {
70
- source: '1',
71
- target: '2',
72
- weight: 0.5,
73
- type: 'semantic',
74
- };
75
- graph.addEdge(edge);
76
- expect(graph['edges'].length).toBe(2);
77
- });
78
- });
79
-
80
- describe('createGraph', () => {
81
- it("chunks and embeddings can't be empty", () => {
82
- const graph = new GraphRAG(3);
83
- const chunks: GraphChunk[] = [];
84
- const embeddings: GraphEmbedding[] = [];
85
- expect(() => graph.createGraph(chunks, embeddings)).toThrowError(
86
- 'Chunks and embeddings arrays must not be empty',
87
- );
88
- });
89
- it('chunks and embeddings must have the same length', () => {
90
- const graph = new GraphRAG(3);
91
- const chunks: GraphChunk[] = [
92
- {
93
- text: 'Chunk 1',
94
- metadata: {},
95
- },
96
- {
97
- text: 'Chunk 2',
98
- metadata: {},
99
- },
100
- ];
101
- const embeddings: GraphEmbedding[] = [
102
- {
103
- vector: [1, 2, 3],
104
- },
105
- ];
106
- expect(() => graph.createGraph(chunks, embeddings)).toThrowError(
107
- 'Chunks and embeddings must have the same length',
108
- );
109
- });
110
- it('should return the top ranked nodes', () => {
111
- const results = [
112
- {
113
- metadata: {
114
- text: 'Chunk 1',
115
- },
116
- vector: [1, 2, 3],
117
- },
118
- {
119
- metadata: {
120
- text: 'Chunk 2',
121
- },
122
- vector: [4, 5, 6],
123
- },
124
- {
125
- metadata: {
126
- text: 'Chunk 3',
127
- },
128
- vector: [7, 8, 9],
129
- },
130
- ];
131
-
132
- const chunks = results.map(result => ({
133
- text: result?.metadata?.text,
134
- metadata: result.metadata,
135
- }));
136
- const embeddings = results.map(result => ({
137
- vector: result.vector,
138
- }));
139
-
140
- const graph = new GraphRAG(3);
141
- graph.createGraph(chunks, embeddings);
142
-
143
- const nodes = graph.getNodes();
144
- expect(nodes.length).toBe(3);
145
- expect(nodes[0]?.id).toBe('0');
146
- expect(nodes[1]?.id).toBe('1');
147
- expect(nodes[2]?.id).toBe('2');
148
-
149
- const edges = graph.getEdges();
150
- expect(edges.length).toBe(6);
151
- });
152
- });
153
-
154
- describe('query', () => {
155
- it("query embedding can't be empty", () => {
156
- const graph = new GraphRAG(3);
157
- const queryEmbedding: number[] = [];
158
- expect(() => graph.query({ query: queryEmbedding, topK: 2, randomWalkSteps: 3, restartProb: 0.1 })).toThrowError(
159
- `Query embedding must have dimension ${3}`,
160
- );
161
- });
162
-
163
- it('topK must be greater than 0', () => {
164
- const graph = new GraphRAG(3);
165
- const queryEmbedding = [1, 2, 3];
166
- const topK = 0;
167
- expect(() => graph.query({ query: queryEmbedding, topK, randomWalkSteps: 3, restartProb: 0.1 })).toThrowError(
168
- 'TopK must be greater than 0',
169
- );
170
- });
171
-
172
- it('randomWalkSteps must be greater than 0', () => {
173
- const graph = new GraphRAG(3);
174
- const queryEmbedding = [1, 2, 3];
175
- const topK = 2;
176
- const randomWalkSteps = 0;
177
- expect(() => graph.query({ query: queryEmbedding, topK, randomWalkSteps, restartProb: 0.1 })).toThrowError(
178
- 'Random walk steps must be greater than 0',
179
- );
180
- });
181
-
182
- it('restartProb must be between 0 and 1', () => {
183
- const graph = new GraphRAG(3);
184
- const queryEmbedding = [1, 2, 3];
185
- const topK = 2;
186
- const randomWalkSteps = 3;
187
- const restartProb = -0.1;
188
- expect(() => graph.query({ query: queryEmbedding, topK, randomWalkSteps, restartProb })).toThrowError(
189
- 'Restart probability must be between 0 and 1',
190
- );
191
- });
192
-
193
- it('should return the top ranked nodes', () => {
194
- const graph = new GraphRAG(3);
195
- const node1: GraphNode = {
196
- id: '1',
197
- content: 'Node 1',
198
- embedding: [1, 2, 3],
199
- };
200
- const node2: GraphNode = {
201
- id: '2',
202
- content: 'Node 2',
203
- embedding: [11, 12, 13],
204
- };
205
- const node3: GraphNode = {
206
- id: '3',
207
- content: 'Node 3',
208
- embedding: [21, 22, 23],
209
- };
210
- graph.addNode(node1);
211
- graph.addNode(node2);
212
- graph.addNode(node3);
213
- graph.addEdge({
214
- source: '1',
215
- target: '2',
216
- weight: 0.5,
217
- type: 'semantic',
218
- });
219
- graph.addEdge({
220
- source: '2',
221
- target: '3',
222
- weight: 0.7,
223
- type: 'semantic',
224
- });
225
-
226
- const queryEmbedding = [15, 16, 17];
227
- const topK = 2;
228
- const randomWalkSteps = 3;
229
- const restartProb = 0.1;
230
- const rerankedResults = graph.query({ query: queryEmbedding, topK, randomWalkSteps, restartProb });
231
-
232
- expect(rerankedResults.length).toBe(2);
233
- });
234
- });
235
- });