Skip to content

framework-support.md

Latest commit

 

History

History
72 lines (54 loc) · 3.07 KB

framework-support.md

File metadata and controls

72 lines (54 loc) · 3.07 KB

Framework Support

tj is OTel-native. Any framework that emits OpenTelemetry spans works automatically — point its OTLP exporter at tj serve and you're done. For everything else, one-line patches exist.

Python provider patches

Intercept at the API level, framework-agnostic:

from tokenjam.sdk.integrations.anthropic import patch_anthropic   # Anthropic — Messages.create + streaming
from tokenjam.sdk.integrations.openai    import patch_openai      # OpenAI — chat completions
from tokenjam.sdk.integrations.gemini    import patch_gemini      # Google Gemini — GenerativeModel
from tokenjam.sdk.integrations.bedrock   import patch_bedrock     # AWS Bedrock — boto3 invoke_model/invoke_agent
from tokenjam.sdk.integrations.litellm   import patch_litellm    # LiteLLM — unified interface for 100+ providers

patch_litellm() covers all providers LiteLLM routes to (OpenAI, Anthropic, Bedrock, Vertex, Cohere, Mistral, Ollama, etc.) with correct per-provider attribution. If you use LiteLLM, you don't need the individual provider patches above.

OpenAI-compatible providers (Groq, Together, Fireworks, xAI, Azure OpenAI) also work via patch_openai(base_url=...) — no separate patches needed.

Python framework patches

Instrument the framework's own tool and LLM abstractions:

from tokenjam.sdk.integrations.langchain         import patch_langchain        # BaseLLM + BaseTool
from tokenjam.sdk.integrations.langgraph         import patch_langgraph        # CompiledGraph
from tokenjam.sdk.integrations.crewai            import patch_crewai           # Task + Agent
from tokenjam.sdk.integrations.autogen           import patch_autogen          # ConversableAgent
from tokenjam.sdk.integrations.llamaindex        import patch_llamaindex       # Native OTel wrapper
from tokenjam.sdk.integrations.openai_agents_sdk import patch_openai_agents   # Native OTel wrapper
from tokenjam.sdk.integrations.nemoclaw          import watch_nemoclaw         # NemoClaw Gateway observer

Zero-code via OTLP

Point any of these frameworks' built-in OTel exporter at tj serve, no integration code required:

Framework OTel support
Claude Code Built-in — setup guide
OpenClaw Built-in (diagnostics-otel plugin) — setup guide
LlamaIndex opentelemetry-instrumentation-llama-index
OpenAI Agents SDK Built-in
Google ADK Built-in
Strands Agent SDK (AWS) Built-in
Haystack Built-in
Pydantic AI Built-in
Semantic Kernel Built-in

TypeScript / Node.js

@tokenjam/sdk provides TjClient and SpanBuilder for sending spans to tj serve from any TypeScript agent:

import { TjClient, SpanBuilder } from "@tokenjam/sdk";

const client = new TjClient({
  baseUrl:      "http://127.0.0.1:7391",
  ingestSecret: process.env.TJ_INGEST_SECRET ?? "",
});

const span = new SpanBuilder("invoke_agent")
  .agentId("my-ts-agent")
  .model("gpt-4o-mini")
  .provider("openai")
  .inputTokens(450)
  .outputTokens(120)
  .build();

await client.send([span]);