truthguard-ai 0.1.2 → 0.1.3

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.
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "truthguard-ai",
3
- "version": "0.1.2",
3
+ "version": "0.1.3",
4
4
  "description": "TruthGuard — Standardized grounding validation for tool-calling AI agents. Detect, diagnose, and prevent grounding failures.",
5
5
  "main": "dist-npm/thin.js",
6
6
  "types": "dist-npm/thin.d.ts",
package/README.npm.md DELETED
@@ -1,144 +0,0 @@
1
- # TruthGuard
2
-
3
- **Grounding validation for tool-calling AI agents.**
4
-
5
- > Detect when an agent's response contradicts the data returned by the tools it called — without LLM-as-judge overhead.
6
-
7
- [![npm version](https://img.shields.io/npm/v/truthguard-ai.svg)](https://www.npmjs.com/package/truthguard-ai)
8
- [![License: MIT](https://img.shields.io/badge/License-MIT-blue.svg)](LICENSE)
9
-
10
- ---
11
-
12
- ## The Problem
13
-
14
- Most "hallucinations" in tool-calling agents are **grounding failures** — the agent calls a tool, gets accurate data, then ignores it, miscalculates, or fabricates from empty results.
15
-
16
- ## The Solution
17
-
18
- TruthGuard extracts factual claims from the agent's response and cross-references them against tool outputs. Deterministic. No LLM calls. Runs in <50ms.
19
-
20
- ```bash
21
- npm install truthguard-ai
22
- ```
23
-
24
- ---
25
-
26
- ## Quick Start
27
-
28
- ### Evaluate a trace
29
-
30
- ```typescript
31
- import { TraceBuilder, GroundingEngine, generateReport } from 'truthguard-ai';
32
-
33
- const trace = new TraceBuilder({ traceId: 'run-001' })
34
- .addUserInput('How many employees are on leave today?')
35
- .addToolCall('getLeaveRecords', { date: '2024-03-15' })
36
- .addToolOutput('getLeaveRecords', [
37
- { employeeId: 'E01', name: 'Ana Jovic', status: 'on_leave' },
38
- { employeeId: 'E02', name: 'Ivan Petrovic', status: 'on_leave' },
39
- ])
40
- .addFinalResponse('There are 3 employees on leave today.')
41
- .build();
42
-
43
- const engine = new GroundingEngine();
44
- const report = engine.evaluate(trace);
45
-
46
- console.log(report.groundingScore); // 0.5
47
- console.log(report.detectedFailures[0]); // { type: 'grounding.data_ignored', severity: 'high' }
48
- ```
49
-
50
- ### SDK Wrappers (OpenAI & Anthropic)
51
-
52
- ```typescript
53
- import { wrapOpenAI } from 'truthguard-ai';
54
- import OpenAI from 'openai';
55
-
56
- const openai = wrapOpenAI(new OpenAI(), {
57
- mode: 'block',
58
- threshold: 0.85,
59
- });
60
- ```
61
-
62
- ```typescript
63
- import { wrapAnthropic } from 'truthguard-ai';
64
- import Anthropic from '@anthropic-ai/sdk';
65
-
66
- const anthropic = wrapAnthropic(new Anthropic(), {
67
- mode: 'warn',
68
- threshold: 0.80,
69
- });
70
- ```
71
-
72
- ### Production Monitoring (Proxy Mode)
73
-
74
- Works with **any language** — PHP, Python, Go, Java, Ruby, C#:
75
-
76
- ```bash
77
- npx truthguard-ai observe --port 3001
78
- ```
79
-
80
- Point your AI SDK base URL to the proxy:
81
- ```
82
- # OpenAI
83
- OPENAI_BASE_URL=http://localhost:3001/proxy/openai
84
-
85
- # Anthropic
86
- ANTHROPIC_BASE_URL=http://localhost:3001/proxy/anthropic
87
- ```
88
-
89
- Your app works identically. TruthGuard evaluates grounding in the background.
90
-
91
- ### CI Quality Gate
92
-
93
- ```bash
94
- npx truthguard-ai run test-cases.jsonl --gate .ai-rcp-gate.yml
95
- ```
96
-
97
- ### MCP Server (VS Code, Cursor)
98
-
99
- Use TruthGuard from your IDE — add to `.vscode/mcp.json`:
100
-
101
- ```json
102
- {
103
- "servers": {
104
- "truthguard": {
105
- "type": "stdio",
106
- "command": "npx",
107
- "args": ["-y", "truthguard-ai", "mcp"]
108
- }
109
- }
110
- }
111
- ```
112
-
113
- ### Express Middleware
114
-
115
- ```typescript
116
- import { groundingMiddleware, FileStore } from 'truthguard-ai';
117
-
118
- app.post('/api/chat', groundingMiddleware({
119
- mode: 'warn',
120
- store: new FileStore('./traces/grounding.jsonl'),
121
- extractTrace: (req, res, body) => body.trace,
122
- }));
123
- ```
124
-
125
- ---
126
-
127
- ## Key Features
128
-
129
- - **30+ deterministic failure detectors** — fabrication from empty results, math errors, ignored tool data, entity mismatches, and more
130
- - **Policy engine** — per-failure enforcement (block / warn / observe)
131
- - **Diagnostic reports** — actionable failure reports with severity levels
132
- - **Baseline regression** — snapshot comparison to catch quality regressions
133
- - **Multi-language support** — works with responses in 13+ languages
134
- - **Configurable thresholds** — tune sensitivity to match your use case
135
-
136
- ---
137
-
138
- ## Documentation
139
-
140
- Full docs, guides, and architecture details: [truthguard.dev](https://truthguard.dev)
141
-
142
- ## License
143
-
144
- MIT
File without changes