> ## Documentation Index
> Fetch the complete documentation index at: https://docs.getpioneer.dev/llms.txt
> Use this file to discover all available pages before exploring further.
# Pioneer
> A local-first AI workspace for desktop, mobile, gateways, tasks, skills, MCP, and local tools.
Pioneer is a local-first AI workspace. Run the assistant on your own machine or on infrastructure you control. The desktop or mobile app is the client. The gateway stores your data, calls model providers, runs tools, manages skills and MCP servers, and keeps threads and tasks alive.
To get started, connect to a gateway, choose or create a workspace, add a model provider, and start a thread. Later, you can add reusable skills, connect MCP servers, attach files as artifacts, enable memory, or schedule repeatable tasks.
Pioneer uses gateway-owned execution with turn permission modes, sandbox policy, resource checks, durable audit events, and keystore-backed secrets. Choose the gateway deliberately: local work runs locally, and remote work runs on the remote gateway.
Agent turns have permission modes. Use **Supervised** when you want Pioneer to ask before sensitive actions, or **Auto-accept edits** when file edits are expected but commands, network, MCP writes, skills, computer use, and subagents should still ask.
## Gateway and clients
Pioneer has four user-facing parts.
**Gateway** runs the work. It may be on your laptop, a workstation, or a server. Provider calls, tool runs, task runs, files, skills, MCP servers, settings, and thread history belong to the gateway you are connected to.
**Desktop app** is the main UI. It connects to gateways, can start and manage a local gateway, and provides screens for workspaces, conversations, provider setup, MCP, skills, settings, and thread history.
**Mobile app** is the iOS and Android client. It connects to existing gateways so you can continue work from a phone or tablet.
**Protocol clients** are custom clients built by teams that want to integrate with Pioneer directly.
For a single-machine setup, install the desktop app and let it start the local gateway for you. On macOS, that means downloading the `.dmg`, moving Pioneer to **Applications**, launching it, and pressing **Start local gateway** when prompted.
For separate environments, install gateways wherever the work should live: a laptop, a workstation, a home server, or a remote machine. Connect to each one from the desktop or mobile app. Separate gateways keep state, settings, tools, and history apart.
Set up the desktop app, a local gateway, or a remote gateway.
Connect a provider and start your first thread.
Understand gateways, workspaces, threads, tools, skills, MCP, and tasks.
Choose when Pioneer should ask before tool actions.
Connect model providers and understand where model settings are saved.
## Multi-agent workflows
Pioneer is designed for more than one long-running chat thread. A task can split work into smaller pieces, give those pieces to subagents, and bring the results back for review.
Each subagent can be scoped independently:
| Scope | What it controls |
| ----------------- | -------------------------------------------------------------------------------------------------------------- |
| Role and identity | Give the agent a role and nickname so its work is understandable in task history. |
| Model choice | Choose the model and provider per agent instead of forcing every subtask through the same model. |
| Context policy | Inherit parent context, pass only recent turns, use a summary, start empty, or provide custom context. |
| Tool policy | Allow or deny tools, choose read-only or write-capable modes, restrict paths, and control network access. |
| Result contract | Ask for text, Markdown, JSON, or artifacts with required outputs. |
| Review policy | Let the parent agent accept a candidate result or request a targeted revision before the result becomes final. |
| Depth and lineage | Track nested agent work through child threads and task lineage so delegated work can be audited and recovered. |
The gateway coordinates this work. One agent can split a job into subtasks, subagents can handle those pieces in separate threads, and the parent can review the results before accepting them. If a result is incomplete, the parent can request a revision with specific feedback.
## Durable agent memory
Pioneer can remember stable facts, preferences, project decisions, recurring instructions, and communication style when memory is enabled.
Memory is different from a long chat transcript. It is meant for durable facts that should be useful later. You can review memory behavior from the desktop settings, and you should treat remembered facts as part of the gateway you are connected to.
For conversation continuity, Pioneer also has thread context: a searchable layer for useful old discussion snippets. That lets the assistant recover prior context without turning every old message into permanent memory.
## Where to start
Start with [Installation](/getting-started/installation), then follow [Quick Start](/getting-started/quickstart). After that, use [Core Concepts](/getting-started/concepts) when a term in the UI is unclear, [Desktop](/desktop/overview) or [Mobile](/mobile/overview) for client-specific behavior, [Providers](/providers/overview) when you want to connect models, [Memory](/desktop/memory) when you want continuity across turns, [Skills](/skills/overview) when you want reusable capabilities, [MCP Servers](/mcp/overview) when you want external tools, and [Tasks](/tasks/overview) when you want repeatable or scheduled work.
If you want the technical version of how Pioneer is built, start with [Architecture Overview](/architecture/overview). The most useful deep dives after this page are [Gateway](/architecture/gateway), [Client Architecture](/architecture/clients), [Agent Loop](/architecture/agent-loop), [Permission System](/architecture/permissions), and [Protocol Layer](/architecture/protocol).