Files
agentlens/packages/sdk-ts/README.md
Vectry f0ce0f7884 feat: SEO improvements and npm publish prep
- Expand sitemap from 2 to 13 URLs (all docs pages)
- Update JSON-LD featureList with Anthropic, OpenCode, TypeScript SDK
- Update llms.txt with docs links, TS SDK, OpenCode plugin sections
- Add READMEs for agentlens-sdk and opencode-agentlens packages
- Add repository, homepage, author, bugs fields to both package.json
2026-02-10 03:43:04 +00:00

125 lines
3.6 KiB
Markdown

# agentlens-sdk
TypeScript SDK for AgentLens — Agent observability that traces decisions, not just API calls.
[![npm version](https://img.shields.io/npm/v/agentlens-sdk.svg)](https://www.npmjs.com/package/agentlens-sdk)
[![license](https://img.shields.io/npm/l/agentlens-sdk.svg)](https://github.com/repi/agentlens/blob/main/LICENSE)
## Install
```bash
npm install agentlens-sdk
```
## Quick Start
```typescript
import { init, TraceBuilder, shutdown } from "agentlens-sdk";
// Initialize the SDK
init({
apiKey: "your-api-key",
endpoint: "https://agentlens.vectry.tech/api",
});
// Create a trace
const trace = new TraceBuilder("agent-run-123", "My Agent Task");
// Add a span (tool call, LLM call, etc.)
trace.addSpan({
name: "search-documents",
type: "tool",
input: { query: "quarterly report" },
output: { results: 5 },
});
// Record a decision point
trace.addDecision({
name: "select-tool",
type: "tool_selection",
options: ["search", "calculate", "summarize"],
selected: "search",
reasoning: "User asked for document lookup",
});
// Finalize and send
await trace.end();
// Flush remaining data before exit
await shutdown();
```
## API Reference
### Core Functions
| Function | Description |
|---|---|
| `init(options)` | Initialize the SDK with your API key and configuration. |
| `shutdown()` | Flush pending data and shut down the transport. |
| `flush()` | Manually flush the current batch without shutting down. |
| `getClient()` | Return the initialized client instance. |
### TraceBuilder
The primary interface for constructing traces.
```typescript
const trace = new TraceBuilder(traceId: string, name: string);
trace.addSpan(span: SpanPayload); // Add a span to the trace
trace.addDecision(decision: DecisionPointPayload); // Record a decision point
trace.end(): Promise<void>; // Finalize and send the trace
```
### Standalone Helpers
| Function | Description |
|---|---|
| `createDecision(decision)` | Create and send a standalone decision point outside a trace. |
## Types
Core payload types used throughout the SDK:
- **`TracePayload`** — Top-level trace structure containing spans and metadata.
- **`SpanPayload`** — Individual unit of work (tool call, LLM request, retrieval, etc.).
- **`DecisionPointPayload`** — A recorded decision: what options existed, what was chosen, and why.
- **`EventPayload`** — Discrete event within a span or trace.
- **`JsonValue`** — Flexible JSON-compatible value type for inputs/outputs.
## Enums
| Enum | Values |
|---|---|
| `TraceStatus` | Status of the overall trace (e.g., running, completed, failed). |
| `SpanType` | Category of span (e.g., tool, llm, retrieval). |
| `SpanStatus` | Status of an individual span. |
| `DecisionType` | Category of decision (e.g., tool_selection, routing). |
| `EventType` | Category of event within a span. |
## Configuration
Pass `InitOptions` to `init()`:
```typescript
init({
apiKey: "your-api-key", // Required. Your AgentLens API key.
endpoint: "https://...", // API endpoint. Defaults to AgentLens cloud.
maxBatchSize: 100, // Max items per batch before auto-flush.
flushInterval: 5000, // Auto-flush interval in milliseconds.
});
```
## Transport
The SDK ships with `BatchTransport`, which batches payloads and flushes them on an interval or when the batch size threshold is reached. This is used internally by `init()` — you typically do not need to instantiate it directly.
## Documentation
Full documentation: [agentlens.vectry.tech/docs/typescript-sdk](https://agentlens.vectry.tech/docs/typescript-sdk)
## License
MIT