AI Decision System
Replayable, offline-verifiable records for AI-assisted go/no-go decisions: experiment promotions, typed model answers and human-approved actions.
翻譯準備中
Track LLM application performance and manage prompts for engineering teams
翻譯準備中
AI 摘要 · README
開發工具Python · TypeScriptGitHub @respanai自行安裝營運者未確認
試用過了嗎? 點一下就好,不需要登入。
尚未偵測到更新。GitHub 發布版本後會顯示在這裡。
個人 GitHub 帳號下公開儲存庫的擁有者可以登入並驗證所有權,自行管理介紹,無需管理員核准。
登入並產生驗證檔案,在 30 分鐘內將其提交至此儲存庫的預設分支。其他分支或分叉儲存庫中的檔案無效。然後在下方檢查檔案,即可無需管理員核准直接管理專案。請保留檔案以維持管理權限。
如需登入,將前往登入頁面。登入後請回到此處,再次點擊所選的驗證按鈕。
安裝型專案無需部署網域也可驗證。個人儲存庫可透過 GitHub 登入驗證,組織儲存庫可使用上方的儲存庫驗證檔案。
正在載入完整 README。 以下為已儲存的摘錄。
Observability, prompt management, and evals for LLM engineering teams.
Respan's library for sending telemetries of LLM applications in OpenLLMetry format.
Replace openai with the provider SDK your application uses. The facade supplies the eligible first-party instrumentation adapters; TypeScript users should not install with --omit=optional because those adapters are optional dependencies of @respan/respan.檢視完整 README ↗Respan's library for sending telemetries of LLM applications in OpenLLMetry format.
pip install respan-ai openai
npm install @respan/respan openai
Replace openai with the provider SDK your application uses. The facade supplies the eligible first-party instrumentation adapters; TypeScript users should not install with --omit=optional because those adapters are optional dependencies of @respan/respan.
| Variable | Required | Description |
|---|---|---|
RESPAN_API_KEY | Yes | Your Respan API key. Authenticates both proxy and tracing. Get it from the platform. |
RESPAN_BASE_URL | No | Defaults to https://api.respan.ai/api. |
The quickstart below routes OpenAI through the Respan gateway, so RESPAN_API_KEY authenticates both inference and telemetry export. When calling a provider directly, configure that provider's credentials and endpoint normally; Respan still uses RESPAN_API_KEY for telemetry.
import os
from openai import OpenAI
from respan import Respan
# Omit `instrumentations` to auto-discover the matching bundled adapter.
respan = Respan()
# Respan API key authenticates both proxy and tracing
respan_api_key = os.environ["RESPAN_API_KEY"]
respan_base_url = os.getenv("RESPAN_BASE_URL", "https://api.respan.ai/api")
client = OpenAI(api_key=respan_api_key, base_url=respan_base_url)
response = client.chat.completions.create(
model="gpt-4.1-nano",
messages=[{"role": "user", "content": "Say hello in three languages."}],
)
print(response.choices[0].message.content)
respan.shutdown()
import OpenAI from "openai";
import { Respan } from "@respan/respan";
const respanBaseURL =
process.env.RESPAN_BASE_URL ?? "https://api.respan.ai/api";
// Omit `instrumentations` to auto-discover the matching bundled adapter.
const respan = new Respan({
apiKey: process.env.RESPAN_API_KEY,
baseURL: respanBaseURL,
});
await respan.initialize();
const client = new OpenAI({
apiKey: process.env.RESPAN_API_KEY,
baseURL: respanBaseURL,
});
const response = await client.chat.completions.create({
model: "gpt-4.1-nano",
messages: [{ role: "user", content: "Say hello in three languages." }],
});
console.log(response.choices[0].message.content);
await respan.shutdown();
See your traces in the Respan platform.
Install the facade and the provider SDK, omit the instrumentations option, and initialize Respan before the first provider call. Missing provider SDKs are skipped without failing startup and are visible through the instrumentation status API.
| Provider | Python SDK | TypeScript SDK |
|---|---|---|
| OpenAI | openai | openai |
| Azure OpenAI | openai | openai |
| Anthropic | anthropic | @anthropic-ai/sdk |
| Vertex AI | google-cloud-aiplatform | @google-cloud/vertexai |
| Google GenAI | google-genai | — |
| AWS Bedrock | boto3 | @aws-sdk/client-bedrock-runtime |
| Cohere | — | cohere-ai |
| Together AI | together | together-ai |
| OpenRouter | — | @openrouter/sdk |
| Writer | — | writer-sdk |
| Ollama | ollama | — |
respan-ai bundles the seven eligible Python adapters. A normal @respan/respan install includes the nine TypeScript adapters as optional dependencies.
Inspect the result after construction in Python with respan.get_auto_instrumentation_status(), or after await respan.initialize() in TypeScript with respan.getInstrumentationStatus().
LLM wrappers, agent frameworks, application frameworks, protocol/tooling integrations, observability bridges, and vector databases remain explicit opt-ins to prevent duplicate spans. Supplying instrumentations selects explicit mode and disables automatic discovery by default.
Python can intentionally combine explicit plugins with direct-SDK automatic discovery by setting is_auto_instrument=True. Use Respan(is_auto_instrument=False) to disable automatic discovery entirely.
In TypeScript, use disabledInstrumentations: ["openrouter"] to disable one automatic adapter, or instrumentations: [] to disable all automatic discovery. TypeScript explicit mode is currently exclusive and cannot be combined with automatic discovery.
Structure traces with @workflow / @task (Python) or withWorkflow / withTask (TypeScript). See the decorators example for details.
Attach customer_identifier, thread_identifier, and metadata to all spans in scope. See the attributes example.
The SDKs batch spans in memory and retry, but never write to disk, so spans still queued during a long outage or a process crash are lost. For long-running agents that must not lose spans, run the packaged Respan Collector next to the application — the upstream OpenTelemetry Collector with a persistent, bounded queue — and point RESPAN_BASE_URL at it (respan collector start or docker compose -f collector/docker-compose.yaml up -d).
Please star us if you found this helpful!