# Little Stack

- [Little Stack documentation](/docs): LittleDB is the product, Little Harness and Little Workflow are the SDK, and workspaces are where an agent's work lives. Choose a piece, set up a project, and see how they fit.
- Get started
  - [Get started](/docs/get-started): Pick the first step that matches what you are building.
  - [Setup](/docs/get-started/setup): Create a Node or Next.js project with a Little Harness agent, or add one to an existing Next.js app, with little setup.
- [How they fit](/docs/how-they-fit): LittleDB is the product, Little Harness and Little Workflow are the SDK, and workspaces are where an agent's work lives. What each owns and how they connect.
- [Releases](/docs/releases): What each Little product has published, what is source-only, and what UNRELEASED means on this site.
- [Glossary](/docs/glossary): Shared terms across Little Harness, Little Workflow, Little Compute and LittleDB, and where each product uses the same word differently.
- [URL migration](/docs/url-migration): Where pages from the separate Little Workflow, Little Harness and Little Compute docs sites live on little-stack.dev.

- **Products**
- Little Harness: Agent runtime · npm 0.2.0-alpha.2
  - [What Little Harness Is](/docs/harness): A zero-config agent runtime for tools, skills, files, traces, and isolated sessions.
  - Guides and API
    - [Guides and API](/docs/harness/v0.1.0-alpha): The Little Harness alpha developer API and local runtime.
    - [Getting Started](/docs/harness/v0.1.0-alpha/getting-started): Scaffold a Little Harness agent workspace and chat with it from the CLI in minutes.
    - [Embedding in Your App](/docs/harness/v0.1.0-alpha/quickstart): Configure and run a Little Harness agent inside your own app — createHarness, streaming, non-chat runs, and sessions.
    - Foundations
      - [Core Mental Model](/docs/harness/v0.1.0-alpha/foundations): The main concepts behind Little Harness: sessions, files, tools, persistence, and traces.
      - [Sessions](/docs/harness/v0.1.0-alpha/foundations/sessions): How Little Harness creates, resumes, and isolates agent sessions.
      - [Files And Artifacts](/docs/harness/v0.1.0-alpha/foundations/files-and-artifacts): The virtual filesystem the runtime exposes to agents — sessions, artifacts, skills, and persistent dirs.
      - [Tools Vs Exports](/docs/harness/v0.1.0-alpha/foundations/tools-vs-exports): How to separate agent capabilities from deliberate app writes.
      - [Persistent Dirs](/docs/harness/v0.1.0-alpha/foundations/persistent-dirs): Cross-session files backed by developer-provided storage logic, and how their writes commit.
      - [Tracing](/docs/harness/v0.1.0-alpha/foundations/tracing): What the host app can inspect after an agent run.
    - Build Agents
      - [Basic Agent](/docs/harness/v0.1.0-alpha/build-agents/basic-agent): The smallest useful Little Harness configuration.
      - [Tools](/docs/harness/v0.1.0-alpha/build-agents/tools): How to expose app capabilities to a Little Harness agent.
      - [Skills](/docs/harness/v0.1.0-alpha/build-agents/skills): How skills teach an agent procedures without bloating every prompt.
      - [Memory](/docs/harness/v0.1.0-alpha/build-agents/memory): Give a Little Harness agent long-term memory as plain files it reads and edits, in one folder or in layers such as company knowledge and per-user preferences.
      - [Message Files](/docs/harness/v0.1.0-alpha/build-agents/message-files): Staging user-uploaded files from UIMessage input into the harness filesystem.
      - [Input Types](/docs/harness/v0.1.0-alpha/build-agents/input-types): Defining reusable non-chat inputs for generateHarness and streamHarness.
      - [Generating And Streaming](/docs/harness/v0.1.0-alpha/build-agents/generating-and-streaming): Using streamHarness and generateHarness in real app routes, jobs, and agent flows.
      - [Agent Folders](/docs/harness/v0.1.0-alpha/build-agents/agent-folders): Keep many agents in one repo, each a self-contained folder, with a CLI to scaffold and test them.
      - [Serving An Agent](/docs/harness/v0.1.0-alpha/build-agents/serving-agents): Run an agent folder as a server with little-harness start, develop it with little-harness dev, and run it on a timer with schedules.
      - [Chat Connectors](/docs/harness/v0.1.0-alpha/build-agents/chat-connectors): Expose an agent folder through Chat SDK platforms or a rich web chat route.
      - [Email Connector](/docs/harness/v0.1.0-alpha/build-agents/email-connector): Give an agent its own email address, so that mail sent to it starts a run and the reply goes back in the same thread.
      - [Outcome Capture](/docs/harness/v0.1.0-alpha/build-agents/outcomes): Record whether a run actually worked — from app code or chat reactions — and read success rates back per prompt.
      - [User Feedback](/docs/harness/v0.1.0-alpha/build-agents/user-feedback): Sending your users' thumbs, ratings, comments and surveys to LittleDB is documented in the LittleDB section. Private source; not installable.
      - [Asking A Person](/docs/harness/v0.1.0-alpha/build-agents/asking-a-person): Let the agent ask a person a question without stopping, and continue the turn when the answer comes.
      - [Triggering a workflow from an agent](/docs/harness/v0.1.0-alpha/build-agents/workflows): Expose a Little Workflow as a typed tool a Little Harness agent can call.
      - [Letting an agent author its own workflow](/docs/harness/v0.1.0-alpha/build-agents/dynamic-workflows): Enable one-shot, model-authored plans a Little Harness agent runs when no configured workflow fits.
      - [Agent Groups](/docs/harness/v0.1.0-alpha/build-agents/agent-groups): Fan out isolated agents through a neutral, production-safe launcher contract.
    - Runtime
      - [Runtime Overview](/docs/harness/v0.1.0-alpha/runtime): The local runtime shape for Little Harness.
      - [Local Host](/docs/harness/v0.1.0-alpha/runtime/local-host): Zero-service local development for Little Harness sessions.
      - [just-bash Runtime](/docs/harness/v0.1.0-alpha/runtime/just-bash): Why the first local Little Harness runtime can build on just-bash.
      - [Execution Environments](/docs/harness/v0.1.0-alpha/runtime/execution-environments): The decoupled sandbox, tool, session, and orchestration ports — execution modes, the subprocess sandbox, custom environments, and the remote session log.
      - [Execution Tiers](/docs/harness/v0.1.0-alpha/runtime/execution-tiers): Classify shell scripts before they run, and escalate the commands just-bash cannot run to a real sandbox.
      - [Isolation And Security](/docs/harness/v0.1.0-alpha/runtime/isolation-and-security): Trust boundaries for sessions, tools, storage, and scripts.
    - Reference
      - [API Reference](/docs/harness/v0.1.0-alpha/reference/api-reference): Alpha API reference for Little Harness.
      - [createHarness()](/docs/harness/v0.1.0-alpha/reference/create-harness): Define a reusable Little Harness runtime surface.
      - [streamHarness()](/docs/harness/v0.1.0-alpha/reference/stream-harness): Stream an interactive Little Harness agent turn.
      - [generateHarness()](/docs/harness/v0.1.0-alpha/reference/generate-harness): Run a Little Harness agent turn and return a completed result.
      - [HarnessSession](/docs/harness/v0.1.0-alpha/reference/harness-session): Session handles, continuation, and out-of-band file staging.
      - [FileWriter](/docs/harness/v0.1.0-alpha/reference/file-writer): Writing files into a Little Harness session from hooks, tools, and host code.
      - [skill()](/docs/harness/v0.1.0-alpha/reference/skill-helper): Proposed helper for adding procedural skill directories to a harness.
      - [Persistent Dir Reference](/docs/harness/v0.1.0-alpha/reference/persistent-dir-reference): Proposed callback shapes for developer-backed persistent directories.
      - [localHost()](/docs/harness/v0.1.0-alpha/reference/local-host): Configure the zero-service local development host.
      - [CLI](/docs/harness/v0.1.0-alpha/reference/cli): Scaffold and interactively test agents, and inspect Local Host sessions and traces.
      - [Events](/docs/harness/v0.1.0-alpha/reference/events): HarnessEvent envelope and trace event vocabulary for sessions, models, tools, artifacts, and commits.
      - [Alpha Migrations](/docs/harness/v0.1.0-alpha/reference/alpha-migrations): Migrate from little-harness 0.2.0-alpha.0 to @little-stack/harness 0.2.0-alpha.2.
      - [Storage Adapters](/docs/harness/v0.1.0-alpha/reference/storage-adapters): The shipped local and customDir adapters, plus a proposed object-store shape.
      - [Using LittleDB](/docs/harness/v0.1.0-alpha/reference/using-littledb): LittleDB's Harness integration moved to the LittleDB section. Private source; not installable.
      - [Glossary](/docs/harness/v0.1.0-alpha/reference/glossary): Short definitions for Little Harness and its Little Compute integration.
    - Cookbook
      - [Next.js useChat Route](/docs/harness/v0.1.0-alpha/cookbook/nextjs-use-chat): Wire streamHarness() into a Next.js App Router route consumed by AI SDK useChat().
      - [Document Research Agent](/docs/harness/v0.1.0-alpha/cookbook/document-research-agent): A recipe for agents that search, inspect, compare, and summarize documents.
      - [Customer Research Agent](/docs/harness/v0.1.0-alpha/cookbook/customer-research-agent): A recipe for account, support, and renewal research agents.
      - [Internal Data Agent](/docs/harness/v0.1.0-alpha/cookbook/internal-data-agent): A recipe for agents that query databases and analyze large results.
      - [gbrain As Memory](/docs/harness/v0.1.0-alpha/cookbook/gbrain-memory): A recipe for keeping Harness memory in a gbrain brain over MCP, with one gbrain source per memory and tokens gbrain scopes to each.
    - Comparisons
      - [Ecosystem Comparison](/docs/harness/v0.1.0-alpha/comparisons): A hub for comparing Little Harness with agent SDKs, frameworks, sandboxes, visual builders, memory layers, and observability tools.
      - [AI SDK Vs Little Harness](/docs/harness/v0.1.0-alpha/comparisons/ai-sdk-vs-little-harness): How Little Harness builds on AI SDK-style streaming without replacing it.
      - [OpenAI Agents SDK Vs Little Harness](/docs/harness/v0.1.0-alpha/comparisons/openai-agents-sdk-vs-little-harness): When to choose OpenAI's agent runtime versus a model-neutral app harness.
      - [LangGraph Vs Little Harness](/docs/harness/v0.1.0-alpha/comparisons/langgraph-vs-little-harness): When to choose LangGraph's durable graph runtime versus Little Harness's managed agent session workspace.
      - [Mastra Vs Little Harness](/docs/harness/v0.1.0-alpha/comparisons/mastra-vs-little-harness): Comparing Mastra's full TypeScript AI framework with Little Harness's narrower runtime primitive.
      - [Other Agent Frameworks Vs Little Harness](/docs/harness/v0.1.0-alpha/comparisons/agent-frameworks-vs-little-harness): Comparing CrewAI, Microsoft Agent Framework, and AutoGen with Little Harness.
      - [Coding Harnesses Vs Little Harness](/docs/harness/v0.1.0-alpha/comparisons/coding-harnesses-vs-little-harness): Comparing OpenHands and OpenClaw with Little Harness's broader product-agent runtime.
      - [Sandboxes Vs Little Harness](/docs/harness/v0.1.0-alpha/comparisons/sandboxes-vs-little-harness): Where sandbox providers, Little Compute, and Little Harness each fit.
      - [Visual Builders Vs Little Harness](/docs/harness/v0.1.0-alpha/comparisons/visual-builders-vs-little-harness): How Dify, Flowise, Langflow, n8n, and Zapier Agents compare with Little Harness.
      - [Memory And Observability Vs Little Harness](/docs/harness/v0.1.0-alpha/comparisons/memory-observability-vs-little-harness): Why memory and tracing tools sit beside Little Harness rather than replacing it.
    - [Roadmap](/docs/harness/v0.1.0-alpha/roadmap): Where Little Harness goes after the alpha — direction, not shipped versions.
- Little Workflow: Durable workflows · npm 0.2.0-alpha.2
  - [What Little Workflow Is](/docs/workflow): The control plane and runtime for AI-generated, durable, inspectable workflows.
  - [Comparisons](/docs/workflow/comparisons): Where Little Workflow fits next to durable execution engines, agent graph frameworks, and low-code automation builders.
  - Guides and API
    - Getting Started
      - [Overview](/docs/workflow/v0.1.0-alpha/getting-started): What Little Workflow is, how it moves from developer intent to durable execution, and where to start.
      - [Quickstart](/docs/workflow/v0.1.0-alpha/getting-started/quickstart): Define a workflow with the ergonomic front door, run it in one call, and inspect the result.
    - Foundations
      - [Core Mental Model](/docs/workflow/v0.1.0-alpha/foundations): How Little Workflow moves from developer intent to harness-based durable execution.
      - [Planner Lifecycle](/docs/workflow/v0.1.0-alpha/foundations/planner-lifecycle): How the Planner harness session turns an OrchestrationRequest into an immutable WorkflowVersion.
    - Authoring
      - [Define a Workflow](/docs/workflow/v0.1.0-alpha/authoring/define-a-workflow): Author a workflow with defineWorkflow — plain input/output schemas, a model or a fixed plan, and inline tools.
      - [Compose With Little Harness](/docs/workflow/v0.1.0-alpha/authoring/compose-with-harness): Hand workflows to a Little Harness agent as typed tools with asHarnessWorkflow — the supported way to compose workflows, replacing the deprecated orchestrator.
      - [Tools](/docs/workflow/v0.1.0-alpha/authoring/tools-and-capabilities): AI SDK tool() as the primitive, MCP integration, and code.run tools proxy.
      - [AI SDK Native](/docs/workflow/v0.1.0-alpha/authoring/ai-sdk-native): How Little Workflow uses AI SDK tool(), model(), and Output.* as first-class primitives.
      - [Memory](/docs/workflow/v0.1.0-alpha/authoring/memory): Persistent Memory Stores and ephemeral Run Scratch — filesystem-shaped storage for agent roles.
      - [Skills](/docs/workflow/v0.1.0-alpha/authoring/skills): Procedurally-disclosed knowledge that teaches agent roles how to handle specific situations.
      - [Orchestrator](/docs/workflow/v0.1.0-alpha/authoring/orchestrating-workflows): Compose multiple workflows dynamically — chains, fan-out, and schema transformation between workflows.
      - [Workflow Config](/docs/workflow/v0.1.0-alpha/authoring/advanced-createlittleworkflow): Every defineWorkflow field — schemas, model slots, planner, fixed plan, worker, tools, and memory.
    - Running
      - [Run A Workflow](/docs/workflow/v0.1.0-alpha/running/run-a-workflow): How to execute a workflow definition against a world, with options, return shape, errors, and streaming.
      - [Local World](/docs/workflow/v0.1.0-alpha/running/local-world): Zero-service local development for durable workflow runs.
      - [Virtual World](/docs/workflow/v0.1.0-alpha/running/virtual-world): Keep workflow events, artifacts and run files in a supplied filesystem without disk.
      - [Inspect Runs](/docs/workflow/v0.1.0-alpha/running/inspect-runs): The event log is the only inspection surface in v0.1.0-alpha.
      - [Durability](/docs/workflow/v0.1.0-alpha/running/replay-and-durability): What survives a process kill in v0.1.0-alpha, and what replay can and cannot do.
      - [Using LittleDB](/docs/workflow/v0.1.0-alpha/running/using-littledb): LittleDB's Workflow integration moved to the LittleDB section. Private source; not installable.
    - Reference
      - [SDK API](/docs/workflow/v0.1.0-alpha/reference/api-reference): The minimal authoring + runtime surface that ships with v0.1.0-alpha.
      - [CLI](/docs/workflow/v0.1.0-alpha/reference/cli): The minimal command surface that ships with v0.1.0-alpha.
      - [Setup Recovery](/docs/workflow/v0.1.0-alpha/reference/setup-recovery): Recover an interrupted little setup run with recoverSetup() and the CLI.
      - [LWIR](/docs/workflow/v0.1.0-alpha/reference/lwir-reference): The Little Workflow Intermediate Representation.
      - [Glossary](/docs/workflow/v0.1.0-alpha/reference/glossary): Short definitions for the primitives that exist in v0.1.0-alpha.
      - [Changelog](/docs/workflow/v0.1.0-alpha/reference/changelog): What changed in the current alpha release of Little Workflow, with a pointer to the full repository changelog.
    - Cookbook
      - [Cookbook](/docs/workflow/v0.1.0-alpha/cookbook): Task-oriented recipes lifted from the real, runnable Little Workflow demos.
      - [Plan once, run many](/docs/workflow/v0.1.0-alpha/cookbook/plan-once-run-many): An orchestrator plans a workflow once and fans it out across many inputs in parallel.
      - [Resume after a crash](/docs/workflow/v0.1.0-alpha/cookbook/resume-after-a-crash): A run survives a mid-flight SIGKILL and resumes deterministically from the durable event log.
    - [Roadmap](/docs/workflow/v0.1.0-alpha/roadmap): Where Little Workflow is headed after the alpha — framed as direction, not shipped versions.
    - Internals
      - [Implementation Details](/docs/workflow/v0.1.0-alpha/internals): How the v0.1.0-alpha SDK actually works, end to end, anchored on the shipped source.
      - [Architecture](/docs/workflow/v0.1.0-alpha/internals/architecture): The layers, module layout, and dependency graph for the alpha SDK — now built around the Harness primitive.
      - [Harness](/docs/workflow/v0.1.0-alpha/internals/harness): The agent-runtime primitive — contract, default workflowHarness, event vocabulary, durability, and plugin model.
      - [Compile Pipeline](/docs/workflow/v0.1.0-alpha/internals/compile-pipeline): How the planner harness session turns a WorkflowDefinition into a locked WorkflowVersion, including WorkflowDefinitionHash and planner memoization.
      - [LWIR Internals](/docs/workflow/v0.1.0-alpha/internals/lwir-internals): The structural validator behind validateLwir() — what it accepts, what it rejects, and what the runtime depends on.
      - [Expressions](/docs/workflow/v0.1.0-alpha/internals/expressions): The deterministic {{ ... }} mini-language used in LWIR and how it is parsed, validated, and evaluated.
      - [Runtime Execution](/docs/workflow/v0.1.0-alpha/internals/runtime-execution): How runWorkflow() drives LWIR steps through the worker harness, with harness session lifecycle, fixer, and two-level durability.
      - [Parallel Execution](/docs/workflow/v0.1.0-alpha/internals/parallel): How bounded parallel branches are scheduled, executed concurrently, persisted, and resumed.
      - [Memory + Scratch](/docs/workflow/v0.1.0-alpha/internals/memory-scratch): Persistent Memory Stores and ephemeral Run Scratch — on-disk layout, mount paths, scoping rules, and bash path enforcement.
      - [Identity Manifests](/docs/workflow/v0.1.0-alpha/internals/identity-manifests): WorkflowDefinitionHash, the three harness manifests, drift detection on resume, and how identity links across a pipeline run.
      - [Fixer](/docs/workflow/v0.1.0-alpha/internals/fixer): Per-step opt-in code repair — when it triggers, what the fixer harness session sees, and how fixes propagate.
      - [Event Log](/docs/workflow/v0.1.0-alpha/internals/event-log): The durability layer. SQLite/WAL-backed event envelopes with two-level vocabulary covering both LWIR steps and harness sessions.
      - [Local World Storage](/docs/workflow/v0.1.0-alpha/internals/local-world-storage): How the alpha world stores events, artifact manifests, blobs, and run locks on the filesystem.
      - [Replay](/docs/workflow/v0.1.0-alpha/internals/replay): How replayRun() rebuilds a run from events — two-level memoization across LWIR steps and harness sessions, with cross-scope sub-run durability.
      - [AI SDK Integration](/docs/workflow/v0.1.0-alpha/internals/ai-sdk-adapter): How the default workflowHarness uses the AI SDK internally — generateText/streamText, Output.* mapping, tool interop, and durability.
      - [SDK Implementation Spec](/docs/workflow/v0.1.0-alpha/internals/sdk-implementation): The build contract for the v0.1.0-alpha Little Workflow SDK.
      - [Security Model](/docs/workflow/v0.1.0-alpha/internals/security-model): Threats and required controls for AI-generated durable workflows.
- Little Compute: Self-hosted agent tasks · unpublished source
  - [What Little Compute Is](/docs/compute): Agent-native scheduling and isolated execution for Little Harness at self-hosted scale.

  - **Operate**
  - [Self-hosting](/docs/compute/self-hosting): CLI-first installation with an always-on control plane and replaceable OCI providers.
  - [CLI Reference](/docs/compute/cli): The default developer commands and retained advanced control surfaces.
  - [Use from Little Harness](/docs/compute/use-from-harness): Give a Little Harness agent a Compute sandbox in one call, and launch Agent Groups on Compute.

  - **Concepts**
  - [Concepts and Terminology](/docs/compute/concepts): Stable names for durable agent work and disposable execution attachments.
  - [Fork and merge](/docs/compute/fork-and-merge): Start many children from the parent's current files, let each work in its own sandbox, and merge their work back into one Project head with per-path policies.
  - [Architecture](/docs/compute/architecture): A production-shaped modular control plane with provider-pluggable isolated execution.
  - [Recovery and safe time](/docs/compute/recovery): Why Little Compute expires authority late rather than early during clock or primary uncertainty.
  - [Providers](/docs/compute/providers): The canonical sandbox boundary, selected substrates, and production promotion gates.
  - [Runtime I/O and Lease admission](/docs/compute/runtime-io): Canonical-first execution capacity, workspace attachment, files, endpoints, and metrics.
  - [Artifact storage](/docs/compute/artifacts): Content-addressed filesystem and S3 storage with immutable retention and generation-fenced garbage collection.

  - **Reference**
  - [HTTP API](/docs/compute/http-api): The bounded, versioned control-plane transport and its safety contract.
  - [Topology credential boundary](/docs/compute/topology-credential-worker): Deploy the unprivileged per-host queue worker and socket-activated root dispatcher.

  - **Status**
  - [Planned enforcement](/docs/compute/planned-enforcement): Egress credential custody and subtree-budget designs; implementation is P2 work.
- LittleDB: Traces, versions and experiments · private source
  - [LittleDB](/docs/littledb): The product of the Little stack. Tracing, versions, experiments and self-improvement for Little Harness and Little Workflow, with one URL and one API key. Private source; not installable.
  - [LittleDB with Little Harness](/docs/littledb/harness): Resolve the version a session runs, send each turn as a trace, record scores, and put agent-made workflows behind review. Private source; not installable.
  - [LittleDB with Little Workflow](/docs/littledb/workflow): A tracing wrapper around any Workflow World that sends each run to LittleDB as a trace. Private source; not installable.
  - [Managed memory](/docs/littledb/memory): Give an agent an org memory everyone shares and a memory for each person, kept and versioned in LittleDB, with review before org memory changes and a Memory tab that shows where each memory came from. Private source; not installable.
  - [User feedback](/docs/littledb/feedback): Send what your users say about an agent's replies (thumbs, ratings, comments, surveys) to LittleDB, where reviews, trials and rollback weigh it next to the judges. Private source; not installable.
  - [Experiments and the eval worker](/docs/littledb/experiments): Datasets with a sealed holdout split, proposals gated on experiments, and the eval worker that runs them inside your app. Private source; not installable.
  - [Guardrails and autonomy](/docs/littledb/guardrails): Decide what LittleDB's improvement loop may ship on its own, write the rules it must keep in your own words, and steer it in plain words with the LittleDB agent. Private source; not installable.