Framework integrations

Availability

Open core · self-host + all Splyntra Cloud plans

Each supported framework has an in-process adapter that Splyntra turns on when you add its name to instrument. Every adapter is a safe no-op if the package isn't installed, so it is harmless to list one you don't use. Only openai, langgraph, openai-agents, and crewai ship pip extras — the rest need no extra.

Instrument names & span mapping

Framework	instrument name	Languages	Span mapping
LangGraph	langgraph	Python + TS	graph run → agent, each node → step
CrewAI	crewai	Python + TS	kickoff → agent, tasks → step, tools → tool_call
OpenAI Agents	openai_agents (or openai-agents)	Python + TS	Runner.run → agent
LlamaIndex	llamaindex	Python	query → agent, retriever → retrieval
MCP	mcp	Python + TS	tools/call → tool_call, per MCP server
Browser Use	browser_use (or browser-use)	Python + TS	Agent.run → agent, each step → step, each browser action (act) → tool_call
Google ADK	google_adk (google-adk)	Python only	agent run → agent
Pydantic AI	pydantic_ai (pydantic-ai)	Python only	agent run → agent

LLM calls made through these frameworks are captured as llm_call spans by the provider adapter (openai, anthropic, ollama) — add the provider name to instrument alongside the framework. See LLM providers.

LangGraph

Python

pip install "splyntra[langgraph,openai]"

from splyntra import Splyntra

Splyntra(
    api_key=os.environ["SPLYNTRA_API_KEY"],
    project="research-agent",
    framework="langgraph",
    instrument=("langgraph", "openai"),
)
# Build and invoke your graph as usual — nodes become step spans.

TypeScript

npm install @splyntra/sdk

import { Splyntra } from "@splyntra/sdk";

new Splyntra({
  apiKey: process.env.SPLYNTRA_API_KEY!,
  project: "research-agent",
  framework: "langgraph",
  instrument: ["langgraph", "openai"],
});
// Build and invoke your graph as usual — nodes become step spans.

CrewAI

Python

pip install "splyntra[crewai,openai]"

from splyntra import Splyntra

Splyntra(
    api_key=os.environ["SPLYNTRA_API_KEY"],
    project="support-crew",
    framework="crewai",
    instrument=("crewai", "openai"),
)
# Build and kickoff() your Crew — tasks become step spans, tools become tool_call spans.

TypeScript

import { Splyntra } from "@splyntra/sdk";

new Splyntra({
  apiKey: process.env.SPLYNTRA_API_KEY!,
  project: "support-crew",
  framework: "crewai",
  instrument: ["crewai", "openai"],
});
// Build and kickoff() your Crew — tasks become step spans, tools become tool_call spans.

OpenAI Agents

Python

pip install "splyntra[openai-agents,openai]"

from splyntra import Splyntra

Splyntra(
    api_key=os.environ["SPLYNTRA_API_KEY"],
    project="triage-agent",
    framework="openai-agents",
    instrument=("openai_agents", "openai"),
)
# Runner.run(...) becomes an agent span.

TypeScript

import { Splyntra } from "@splyntra/sdk";

new Splyntra({
  apiKey: process.env.SPLYNTRA_API_KEY!,
  project: "triage-agent",
  framework: "openai-agents",
  instrument: ["openai-agents", "openai"],
});
// Runner.run(...) becomes an agent span.

LlamaIndex

Python only.

pip install splyntra

from splyntra import Splyntra

Splyntra(
    api_key=os.environ["SPLYNTRA_API_KEY"],
    project="rag-app",
    framework="llamaindex",
    instrument=("llamaindex", "openai"),
)
# query() becomes an agent span; the retriever emits a retrieval span.

MCP (Model Context Protocol)

Python

from splyntra import Splyntra

Splyntra(
    api_key=os.environ["SPLYNTRA_API_KEY"],
    project="mcp-agent",
    instrument=("mcp", "openai"),
)
# Each tools/call becomes a tool_call span, attributed to its MCP server.

TypeScript

import { Splyntra } from "@splyntra/sdk";

new Splyntra({
  apiKey: process.env.SPLYNTRA_API_KEY!,
  project: "mcp-agent",
  instrument: ["mcp", "openai"],
});
// Each tools/call becomes a tool_call span, attributed to its MCP server.

Per-server MCP metrics surface on the MCP Servers screen.

Browser Use

Browser Use drives autonomous web-browsing agents. Both the Python library and the TypeScript port (npm browser-use) are supported; no extra is required — install browser-use alongside the SDK.

Python

pip install splyntra browser-use

from splyntra import Splyntra

Splyntra(
    api_key=os.environ["SPLYNTRA_API_KEY"],
    project="browser-agent",
    framework="browser-use",
    instrument=("browser-use", "openai"),
)
# Run your Agent as usual — agent.run() is the root agent span, each step a step
# span, and every browser action (navigate, click, type, extract, …) a tool_call.

TypeScript

npm install @splyntra/sdk browser-use

import { Splyntra } from "@splyntra/sdk";

new Splyntra({
  apiKey: process.env.SPLYNTRA_API_KEY!,
  project: "browser-agent",
  framework: "browser-use",
  instrument: ["browser-use", "openai"],
});
// Agent.run() → agent span, each step → step span, each browser action → tool_call.

The adapter tracks Browser Use's Controller/Tools action executor across versions (the executor class was renamed as the API evolved), so browser-action tool_call spans keep flowing on both the legacy and current API. Add openai (or your provider) to instrument to capture the agent's vision/reasoning LLM calls as llm_call spans with token cost.

See the runnable Browser agent example for a full walkthrough with URL governance, indirect-prompt-injection defense, and audit logging.

Google ADK

Python only.

from splyntra import Splyntra

Splyntra(
    api_key=os.environ["SPLYNTRA_API_KEY"],
    project="adk-agent",
    instrument=("google_adk",),
)

Pydantic AI

Python only.

from splyntra import Splyntra

Splyntra(
    api_key=os.environ["SPLYNTRA_API_KEY"],
    project="pydantic-agent",
    instrument=("pydantic_ai", "openai"),
)

Next steps

• LLM providers — provider adapters and OpenAI-compatible endpoints.
• Manual instrumentation — trace your own agent, tool, and LLM functions when there is no adapter.
• Connect an agent — generate the exact snippet for your stack.