# Fused

> Fused isan integration gateway for apps and agents. Fused offers you a central integration gateway for your internal and external services you depend on

- [Introduction](https://usefused.com/docs/introduction.md): One controlled path from your applications and agents to every service they depend on.
- [Get an Engine running](https://usefused.com/docs/deploy-an-engine.md): Stand up the runtime everything else in Fused talks to.
- [Quickstart](https://usefused.com/docs/quickstart.md): Install the CLI and ship a working SDK through one guided setup.
- [Enable a service](https://usefused.com/docs/workspace/enable-a-service.md): Make a service available to every SDK and MCP server in your workspace.
- [Use a Managed service](https://usefused.com/docs/workspace/use-managed-service.md): Connect users through a Fused Managed App without storing your own OAuth client pair.
- [Bring your own API in](https://usefused.com/docs/workspace/bring-your-own-api.md): Turn a spec, a GraphQL endpoint, or a documentation page into a versioned service.
- [Build an SDK](https://usefused.com/docs/app/build-an-sdk.md): Turn a set of approved operations into a typed package your product can import.
- [Create or update from a prompt](https://usefused.com/docs/app/from-a-prompt.md): Describe the capabilities you need, review exact operations, and publish an SDK or MCP version.
- [Use it in your app](https://usefused.com/docs/app/use-in-your-app.md): Install the generated package and stand up a client your application can hold.
- [Run an operation](https://usefused.com/docs/app/run-an-operation.md): Call an approved operation from your application and handle what comes back.
- [Authenticate into Fused](https://usefused.com/docs/app/fused-auth-client.md): Use the built-in Fused Auth client to sign a user into Fused with OAuth and act on their behalf.
- [Manage MCP servers programmatically](https://usefused.com/docs/app/fused-admin-client.md): Use the built-in Fused Admin client to list and deploy MCP servers and mint execution tokens with an OAuth access token.
- [Deploy an MCP server](https://usefused.com/docs/mcp/deploy-a-server.md): Stand up one Engine-hosted MCP server that reaches every service an agent is allowed to use.
- [Receive MCP events](https://usefused.com/docs/mcp/receive-events.md): Expose selected provider webhooks as MCP resources, listen for updates, and read the latest event.
- [Connect an MCP client](https://usefused.com/docs/mcp/use-the-server.md): Place the stable Fused MCP URL and an agent token in your MCP client.
- [Handle provider OAuth](https://usefused.com/docs/mcp/use-the-server/provider-oauth.md): Advertise URL elicitation so your MCP client can let users connect provider accounts when an operation needs them.
- [Request authentication](https://usefused.com/docs/mcp/use-the-server/provider-oauth/request-authentication.md): Advertise URL elicitation and present the authorization link returned by Fused.
- [Handle completion and resume the agent](https://usefused.com/docs/mcp/use-the-server/provider-oauth/handle-completion.md): Receive the OAuth completion notification, resume the pending task, and prevent duplicate execution.
- [Store credentials](https://usefused.com/docs/bucket/store-credentials.md): Put a provider credential somewhere your apps can use it without ever seeing it.
- [Connect a user's account](https://usefused.com/docs/bucket/connect-user-accounts.md): Run OAuth once, per end user, and let the Engine keep the tokens working.
- [Control the workflow](https://usefused.com/docs/advanced/control-the-workflow.md): Use explicit validation, plan, apply, receipts, and structured output when the guided path is not enough.
- [Manage versions and deprecation](https://usefused.com/docs/workspace/manage-versions.md): Run several versions of a service at once, and retire one without breaking what already shipped.
- [Handle provider limits and failures](https://usefused.com/docs/workspace/handle-provider-limits.md): Rate limits, retries, and pagination handled once in the runtime instead of in every feature.
- [Share access with teams](https://usefused.com/docs/workspace/share-access.md): Give a team exactly the services, buckets, and apps it needs — and nothing else.
- [Ship a new version](https://usefused.com/docs/app/ship-a-new-version.md): Change what an SDK exposes without rewriting the contract of software already in production.
- [Receive SDK events](https://usefused.com/docs/app/receive-events.md): Handle provider webhooks and Fused auth lifecycle events through one durable SDK receiver.
- [Issue execution tokens](https://usefused.com/docs/app/issue-tokens.md): Give a running application a credential that calls your SDK — and nothing else.
- [Test from the CLI](https://usefused.com/docs/app/testing/from-the-cli.md): Prove credentials, scope, and routing from a terminal or a CI job — without writing application code.
- [Call over REST](https://usefused.com/docs/app/testing/over-rest.md): One HTTP endpoint, for languages with no generated package.
- [Declare a unified operation](https://usefused.com/docs/app/unified/declare.md): Turn several provider calls into one SDK or MCP operation, with dependencies and rollback.
- [Call a unified operation](https://usefused.com/docs/app/unified/call.md): Choose targets, route each service, and read what comes back.
- [Create a token for an agent](https://usefused.com/docs/mcp/agent-tokens.md): Issue a scoped MCP token and optionally bind it to connected users.
- [Tokens for OAuth services](https://usefused.com/docs/mcp/oauth-tokens.md): Give an agent access to a provider that acts on behalf of individual users.
- [Share a bucket](https://usefused.com/docs/bucket/share-a-bucket.md): Let another team build on your credentials without being able to read them.
- [Know when something changes](https://usefused.com/docs/notifications.md): Provider drift, version changes, and the records of your own destructive decisions.
- [Build agents for production.](https://usefused.com/docs/harnest/overview.md): Harnest is a production harness for Google ADK and LangGraph agents.
- [Install](https://usefused.com/docs/harnest/get-started/install.md): Install the Harnest CLI and verify its managed runtime.
- [Initialize a project](https://usefused.com/docs/harnest/get-started/initialize.md): Create an agent, connect your model, and try it locally.
- [Migrate a project](https://usefused.com/docs/harnest/get-started/migrate.md): Bring an existing agent into Harnest, upgrade a project, or switch frameworks.
- [Run an agent locally](https://usefused.com/docs/harnest/get-started/run.md): Invoke the compiled root agent without starting an HTTP server.
- [Serve a project](https://usefused.com/docs/harnest/get-started/serve.md): Launch the neutral server and development playground.
- [Agent concepts](https://usefused.com/docs/harnest/build/agent-concepts.md): Understand the concepts Harnest composes and where each concept is authored.
- [Project configuration](https://usefused.com/docs/harnest/build/project-configuration.md): Configure the agent, framework, server, discovery metadata, and dependencies.
- [Agents and graphs](https://usefused.com/docs/harnest/build/agents-and-graphs.md): Author managed agents, portable graphs, structured results, and native targets.
- [Python imports](https://usefused.com/docs/harnest/build/python-imports.md): Import authoring contracts from their Harnest domain.
- [Configure a model](https://usefused.com/docs/harnest/build/models-and-libraries/configure-a-model.md): Choose an OpenAI-compatible model and endpoint, then supply credentials for local development, CI, deployment, and evaluations.
- [Typed decisions](https://usefused.com/docs/harnest/build/models-and-libraries/typed-decisions.md): Give agents typed judgments with custom providers, explicit routing rules, and offline tests.
- [Use Jev for decisions](https://usefused.com/docs/harnest/build/models-and-libraries/typed-decisions/jev.md): Route support tickets with Jev through Harnest's custom-provider API.
- [Decision provider reference](https://usefused.com/docs/harnest/build/models-and-libraries/typed-decisions/providers.md): Implement a custom provider and configure decision validation, policies, failures, and telemetry.
- [Data models and shared libraries](https://usefused.com/docs/harnest/build/data-models-and-shared-libraries.md): Share Pydantic contracts and ordinary Python implementation across an agent bundle.
- [Accept multimodal data](https://usefused.com/docs/harnest/build/models-and-libraries/typed-multimodal-contracts.md): Validate image, audio, video, file, text, and custom data across agents and tools.
- [Store and retrieve media](https://usefused.com/docs/harnest/build/models-and-libraries/store-and-retrieve-media.md): Select a named asset store, persist media, and retrieve it through the invocation context.
- [Token utilisation](https://usefused.com/docs/harnest/build/models-and-libraries/token-utilisation.md): Observe token usage, set budgets, and choose how managed agents reduce model context.
- [Agent Tools](https://usefused.com/docs/harnest/build/agent-tools.md): Choose server tools, client-hosted tools, MCP tools, and approval boundaries.
- [Server tools](https://usefused.com/docs/harnest/build/agent-tools/server-tools.md): Run typed, validated capabilities inside the agent server.
- [Client-hosted tools](https://usefused.com/docs/harnest/build/agent-tools/client-tools.md): Declare typed work that runs in a browser, desktop, or mobile client.
- [Human approvals](https://usefused.com/docs/harnest/build/agent-tools/human-approvals.md): Require permission for every tool call or for one operation found risky at runtime.
- [SubAgents](https://usefused.com/docs/harnest/build/subagents.md): Choose flat, folder-owned, portable, or native SubAgent composition.
- [Flat SubAgents](https://usefused.com/docs/harnest/build/subagents/flat-subagents.md): Add a SubAgent definition that does not need private instructions or capabilities.
- [Folder-owned SubAgents](https://usefused.com/docs/harnest/build/subagents/folder-subagents.md): Give a SubAgent its own instructions, tools, MCP connections, skills, and sandbox.
- [SubAgent framework behavior](https://usefused.com/docs/harnest/build/subagents/framework-behavior.md): Choose portable graph composition, recursive ADK discovery, or an advanced child target.
- [MCP Client](https://usefused.com/docs/harnest/build/mcp-client.md): Discover tools, resources, and prompts; subscribe to retained resource updates.
- [Connect to OpenAPI at runtime](https://usefused.com/docs/harnest/build/openapi-mcp.md): Use MCPClient.from_openapi to turn an OpenAPI specification into a runtime MCP connection.
- [Compose APIs with Fused](https://usefused.com/docs/harnest/build/fused-mcp.md): Combine multiple OpenAPI specifications into one Fused MCP server and connect through a standard Harnest MCP Client.
- [Connect a Slack channel through Fused](https://usefused.com/docs/harnest/build/channels.md): Receive Slack mentions through a shared Fused connection and deliver threaded agent replies.
- [Agent plugins](https://usefused.com/docs/harnest/build/agent-plugins.md): Add portable Agent Plugins containing skills, MCP servers, or both.
- [Company project packs](https://usefused.com/docs/harnest/build/project-packs.md): Build an internal agent product with company scaffolds, typed settings, and versioned migrations.
- [Create a project pack](https://usefused.com/docs/harnest/build/project-packs/create.md): Declare a company pack with typed input options and a validated configuration file.
- [Initialise a project](https://usefused.com/docs/harnest/build/project-packs/initialize.md): Use the initialize decorator to add company defaults to a new Harnest project.
- [Migrate company configuration](https://usefused.com/docs/harnest/build/project-packs/migrations.md): Use consecutive versioned migrations to evolve existing agents without rerunning initialization.
- [Declare project changes](https://usefused.com/docs/harnest/build/project-packs/changes.md): Build ChangePlan operations with explicit ownership and conflict behavior.
- [Build a company CLI](https://usefused.com/docs/harnest/build/project-packs/cli.md): Expose typed init and upgrade commands and register additional company commands.
- [Review plans and recover changes](https://usefused.com/docs/harnest/build/project-packs/plans-and-recovery.md): Use the planning API, track company schema versions, and handle conflicts or interrupted applies.
- [Harnest Extensions](https://usefused.com/docs/harnest/build/extensions.md): Follow the Harnest Extension process from boundary selection through authoring, lifecycle integration, and application use.
- [Create a Harnest Extension](https://usefused.com/docs/harnest/build/extensions/create.md): Author the Harnest Extension manifest, singleton, typed invocation context, and dependency contract.
- [Harnest Extension lifecycle](https://usefused.com/docs/harnest/build/extensions/lifecycle.md): Connect extension startup, invocation context, hooks, storage, telemetry, and auditing to Harnest lifecycle.
- [Use a Harnest Extension](https://usefused.com/docs/harnest/build/extensions/use.md): Discover a Harnest Extension, consume its compiler-owned namespace, and keep access invocation-scoped.
- [Docker extension](https://usefused.com/docs/harnest/build/extensions/official/docker.md): Run framework-neutral Python sandbox workloads through the Fused-maintained Docker Harnest Extension.
- [Hatchet extension](https://usefused.com/docs/harnest/build/extensions/official/hatchet.md): Submit and recover durable external workflows through the Fused-maintained Hatchet Harnest Extension.
- [RAG extension](https://usefused.com/docs/harnest/build/extensions/official/rag.md): Build tenant-scoped PostgreSQL RAG behind an extensible, typed backend contract.
- [Tasks](https://usefused.com/docs/harnest/build/tasks.md): Create retryable application work, defer it from tools, and run it on a schedule.
- [Create and run Tasks](https://usefused.com/docs/harnest/build/queued-tasks.md): Define, defer, inspect, cancel, and await retryable application work.
- [Cron schedules](https://usefused.com/docs/harnest/build/scheduled-tasks.md): Declare fixed UTC task schedules in the cron folder.
- [Dynamic schedules](https://usefused.com/docs/harnest/build/dynamic-schedules.md): Create and control user-owned recurring tasks from managed code.
- [Agent Skills](https://usefused.com/docs/harnest/build/agent-skills.md): Load static or dynamic task guidance without redeploying the agent.
- [Manage skill caching](https://usefused.com/docs/harnest/build/agent-skills/caching.md): Cache dynamic skill catalogs and content while preserving authorization and version consistency.
- [Sandboxing](https://usefused.com/docs/harnest/build/sandboxing.md): Assign named Python execution providers to agents, then call them from your authored tools on ADK or LangGraph.
- [Testing and compilation](https://usefused.com/docs/harnest/build/testing-and-compilation.md): Run unit tests, smoke tests, evals, CI checks, and compile standalone artifacts.
- [Evaluations](https://usefused.com/docs/harnest/build/evaluations.md): Choose reference, judge, safety, tool, multi-turn, simulation, or custom evaluations for ADK and LangGraph agents.
- [Eval sets and configuration](https://usefused.com/docs/harnest/build/evaluations/eval-sets-and-configuration.md): Author validated static conversations, expected behavior, session input, rubrics, and shared evaluation criteria.
- [Metrics reference](https://usefused.com/docs/harnest/build/evaluations/metrics.md): Compare every built-in evaluation metric, its score range, required evidence, criterion shape, and service dependency.
- [Simulations and live evaluation](https://usefused.com/docs/harnest/build/evaluations/simulations-and-live-evaluation.md): Evaluate generated text or audio user turns and run ADK agents through bidirectional live inference.
- [Create custom evals](https://usefused.com/docs/harnest/build/evaluations/create-custom-evals.md): Create, unit-test, register, and run a custom Python scorer for an ADK or LangGraph agent.
- [Custom metric API](https://usefused.com/docs/harnest/build/evaluations/custom-metrics.md): Use Harnest scoring helpers or native ADK metrics to evaluate agents with your own rules and hosted services.
- [Run and inspect evaluation results](https://usefused.com/docs/harnest/build/evaluations/run-and-inspect-results.md): Run every eval suite, save the versioned EvalRunResult JSON, and keep sensitive diagnostics safe in CI.
- [ADK and LangGraph](https://usefused.com/docs/harnest/runtime/adk-and-langgraph.md): Choose a framework, understand portability, and migrate between ADK and LangGraph.
- [Lifecycle](https://usefused.com/docs/harnest/runtime/lifecycle.md): Add application resources, invocation hooks, output policy, and native integrations.
- [Application resources](https://usefused.com/docs/harnest/runtime/lifecycle/application-resources.md): Start dependencies once, publish them to agent code, and close them safely.
- [Invocation hooks](https://usefused.com/docs/harnest/runtime/lifecycle/invocation-hooks.md): Observe or transform agent, model, and tool execution in a portable way.
- [Output policy](https://usefused.com/docs/harnest/runtime/lifecycle/output-policy.md): Control SubAgent narration, tool activity, decision results, provider reasoning, and metadata disclosure.
- [Native integrations](https://usefused.com/docs/harnest/runtime/lifecycle/native-integrations.md): Add ADK plugins or LangGraph middleware without hiding the portability boundary.
- [Authentication and credentials](https://usefused.com/docs/harnest/runtime/authentication-and-credentials.md): Authenticate callers and resolve downstream credentials without exposing secrets.
- [Agent Runtime Principals](https://usefused.com/docs/harnest/runtime/agent-runtime-principals.md): Limit the Harnest capabilities available to one agent invocation.
- [Runtime storage](https://usefused.com/docs/harnest/runtime/checkpoints-and-storage.md): Choose storage authorities for committed sessions, in-progress checkpoints, media, and application data.
- [Task and cron storage](https://usefused.com/docs/harnest/runtime/task-storage.md): Use PostgreSQL, Redis, or a custom database for Harnest-owned durable workers and schedules.
- [Long-term memory](https://usefused.com/docs/harnest/runtime/long-term-memory.md): Explicitly save and retrieve user-scoped memories across sessions using PostgreSQL, Redis, or your own provider.
- [Sessions and application storage](https://usefused.com/docs/harnest/runtime/session-storage.md): Persist committed conversation history and session-scoped application data without mixing them with checkpoints.
- [Checkpoints](https://usefused.com/docs/harnest/runtime/checkpoints.md): Persist private in-progress execution state for recovery, durable waits, and multiple replicas.
- [Durable execution](https://usefused.com/docs/harnest/runtime/durable-execution.md): Suspend managed tools and resume the same agent session from any replica.
- [Telemetry](https://usefused.com/docs/harnest/runtime/telemetry.md): Add logs and traces, then choose one or multiple telemetry sinks.
- [Configure one telemetry sink](https://usefused.com/docs/harnest/runtime/telemetry/single-sink.md): Send Harnest traces and logs to one OTLP collector.
- [Configure multiple telemetry sinks](https://usefused.com/docs/harnest/runtime/telemetry/multiple-sinks.md): Fan out Harnest traces and logs to multiple OTLP destinations.
- [Threadify](https://usefused.com/docs/harnest/runtime/telemetry/threadify.md): Opt into session-linked business telemetry without exporting framework noise.
- [AG-UI](https://usefused.com/docs/harnest/runtime/ag-ui.md): Connect an agent-facing UI to Harnest through streaming text, shared state, browser tools, and human approval.
- [Connect an AG-UI client](https://usefused.com/docs/harnest/runtime/ag-ui/connect.md): Serve a Harnest agent, create a thread, and stream turns through the AG-UI endpoint.
- [Share state and context](https://usefused.com/docs/harnest/runtime/ag-ui/state-and-context.md): Send session-state patches and application context, and receive authored agent state through AG-UI.
- [Handle AG-UI interactions](https://usefused.com/docs/harnest/runtime/ag-ui/interactions.md): Resume approvals and browser tools, and recover the status of external waits.
- [Try the CopilotKit demo](https://usefused.com/docs/harnest/runtime/ag-ui/copilotkit.md): Run a React UI against Harnest agents on Google ADK and LangGraph, then verify the full interaction flow.
- [Agent2Agent (A2A)](https://usefused.com/docs/harnest/runtime/a2a.md): Expose a Harnest agent through A2A and call remote A2A agents through a bounded client.
- [Serve an agent through A2A](https://usefused.com/docs/harnest/runtime/a2a/serve.md): Declare an A2A interface and use Harnest-owned execution, identity, and durable Task storage.
- [Call a remote A2A agent](https://usefused.com/docs/harnest/runtime/a2a/client.md): Use the lazy A2A client directly or compose a remote agent as a graph node or Agent Tool.
- [Serving agents](https://usefused.com/docs/harnest/runtime/serving.md): Run the neutral API, event streams, approvals, client tools, and playground.
- [Server configuration](https://usefused.com/docs/harnest/runtime/serving/server-configuration.md): Override standalone server defaults in your project config.
- [Development reload](https://usefused.com/docs/harnest/runtime/serving/development-reload.md): Recompile authored changes and replace the local agent process without mutating a running graph.
- [Neutral API](https://usefused.com/docs/harnest/runtime/serving/neutral-api.md): Use portable sessions, responses, SSE events, WebSockets, and the playground.
- [Custom HTTP endpoints](https://usefused.com/docs/harnest/runtime/custom-http-endpoints.md): Add application-specific FastAPI endpoints that invoke your Harnest agent.
- [Approvals and client tools](https://usefused.com/docs/harnest/runtime/serving/approvals-and-client-tools.md): Handle required actions and resume suspended invocations over JSON, SSE, or WebSocket.
- [Production deployment](https://usefused.com/docs/harnest/runtime/serving/production.md): Add the security, durability, and operations missing from the standalone launcher.
- [Provision agents and services](https://usefused.com/docs/harnest/runtime/serving/provisioning.md): Deploy agent images and their dependencies locally or to K3s, and connect externally managed databases.
- [Architecture](https://usefused.com/docs/harnest/reference/architecture.md): Understand how Harnest turns agent source into a framework-specific, portable server.
- [Lifecycle reference](https://usefused.com/docs/harnest/reference/lifecycle.md): Find each lifecycle decorator, its allowed count, and its detailed guide.
- [Installation and releases](https://usefused.com/docs/harnest/reference/installation-and-releases.md): Install, pin, verify, upgrade, and publish Harnest releases.
- [Development standards](https://usefused.com/docs/harnest/reference/development-standards.md): Follow Harnest source-quality, testing, complexity, and audit requirements.