Integrations

Since: 1.9.0

Use Gavio as the embedded runtime governance layer beside the tools teams already use.

Application
  -> Gavio embedded runtime
  -> gateway / provider / framework
  -> observability, eval, and security tools through runtime events

Tool	Category	Let that tool own	Let Gavio own	Export
LiteLLM	gateway	Proxy, virtual keys, provider routing	App PII/policy, audit hashes, cost labels	JSONL, OTel
Portkey	gateway	Gateway config, org controls, routing	Embedded policy, interceptor facts, audit	JSONL, OTel
Helicone	gateway observability	Gateway analytics and prompt workflows	Local runtime controls and privacy-safe labels	JSONL
Langfuse	observability	Traces, prompts, eval datasets, review loops	Runtime facts and policy/audit context	JSONL
OpenLIT	observability	OTel-native dashboards and APM correlation	Runtime event source and span attributes	OTel
promptfoo	eval	Eval suites, red-team tests, CI gates	Runtime assertions and safe eval reports	JSONL
LangChain	framework	Chains, agents, tools, memory	Governed model calls and tool validation	JSONL, OTel
LangGraph	framework	Graph state, nodes, checkpoints	Per-node labels and replay evidence	JSONL, OTel
Vercel AI SDK	framework	Streaming UX, server actions, provider APIs	Server-side governance and route labels	JSONL, OTel
OpenAI SDK	provider SDK	Provider-specific APIs, files, assistants	Governed chat shim, policy, audit/export	JSONL, OTel

Use the SDK catalog helpers to keep labels consistent:

from gavio import integration_metadata

metadata = integration_metadata(
    "litellm",
    tenant="acme",
    feature="support-chat",
    environment="prod",
)

import { integrationMetadata } from "gavio/integrations"

const metadata = integrationMetadata("openlit", {
  tenant: "acme",
  feature: "support-chat",
  environment: "prod",
})

import io.gavio.integrations.IntegrationCatalog;

var metadata = IntegrationCatalog.metadata(
    "langchain",
    Map.of("tenant", "acme", "feature", "support-chat", "environment", "prod"));

These scalar labels flow into runtime events and cost dimensions without exporting raw prompt or response text.

Ecosystem trust package

Since: 2.7.0

The Ecosystem Trust Package adds a shared conformance vector and generated compatibility matrix for the integration catalog. The matrix is generated from the catalog vector, adapter payload vector, and ecosystem trust vector, then checked by the SDK tests:

node scripts/gen-ecosystem-trust-matrix.mjs --check

The generated artifact is docs/integrations/compatibility-matrix.json. It records the conformance-tested trust level, metadata-only privacy boundary, adapter-payload coverage, and the offline sample production app that exercises each integration.

Adapter payloads

Since: 2.5.0

Adapter payload helpers build metadata-only fragments for ecosystem SDK calls, configs, callbacks, or telemetry wrappers. They do not import external SDKs.

Tool	Payload target
LiteLLM	completion kwargs metadata and trace headers
promptfoo	default test metadata, Gavio vars, runtime assertions
Langfuse	trace and generation metadata
OpenLIT	OTel/OpenLIT span attributes
LangChain	RunnableConfig metadata and tags
LangGraph	RunnableConfig metadata, tags, and configurable ids
Vercel AI SDK	request headers and experimental telemetry metadata

All adapter payloads share this envelope:

{
  "schemaVersion": "gavio.integration-adapter.v1",
  "adapter": "langfuse",
  "target": "langfuse",
  "kind": "observability",
  "payload": {}
}

Content-bearing metadata fields such as messages, content, diff, prompt, response, output, renderedPrompt, and rendered_prompt are replaced with SHA-256 hash fields. Runtime-event source content is not copied into the adapter summary.

from gavio import integration_adapter_payload

payload = integration_adapter_payload(
    "langfuse",
    {"traceId": "trace_123", "data": {"status": "ok", "provider": "openai"}},
    metadata={"tenant": "acme", "prompt": "raw text"},
)

import { integrationAdapterPayload } from "gavio/integrations"

const payload = integrationAdapterPayload(
  "vercel-ai-sdk",
  { traceId: "trace_123", data: { status: "ok", provider: "openai" } },
  { metadata: { tenant: "acme", prompt: "raw text" } },
)

import io.gavio.integrations.IntegrationAdapters;

var payload = IntegrationAdapters.payload(
    "langchain",
    Map.of("traceId", "trace_123", "data", Map.of("status", "ok", "provider", "openai")),
    Map.of("tenant", "acme", "prompt", "raw text"));