Skip to main content
Pioneer is built around a persistent gateway, a native desktop app, durable threads, task automation, MCP servers, skills, and real local tools. These are the product highlights that matter before you install it.

Gateway-Centered Design

All important work happens in the gateway: workspaces, threads, turns, tools, MCP, skills, tasks, provider settings, auth, and durable storage.

Local Or Remote Deployment

Run a gateway on your personal computer for a local assistant, or host gateways on separate servers for work, study, home, or other isolated environments.

One Desktop, Many Gateways

Connect the desktop app to any number of gateways and switch between them from one native client.

Workspace Management

Create, switch, and rename workspaces inside a gateway. Each workspace keeps its own threads, provider keys, MCP servers, skills, tasks, and artifacts.

Thread Modes

Use Chat mode for direct conversations and Agent mode when the thread should plan, use tools, and work through multi-step tasks.

Thread Tree AGENTS.md

Create AGENTS.md at the thread tree root or inside any thread folder to keep persistent agent instructions close to the work they affect. Threads inherit the nearest active file from their folder ancestry.

Multi-Agent Workflows

The gateway can automatically fan work out to subagents with their own prompts, roles, models, context policies, tool policies, result contracts, and child threads. The parent agent can review each subagent result, accept it, or request a revision with concrete feedback.

Agent Memory

In agent mode, Pioneer can recall durable facts, preferences, project decisions, and recurring instructions across turns when memory is enabled. Memory combines prompt policy, model-visible memory tools, proactive post-turn extraction, service-owned dedupe, and memvid-backed search capsules. It is not full transcript recall, and it does not guarantee every useful fact will be saved.

Bring Your Own Model

Pioneer includes providers for OpenAI, Anthropic, OpenRouter, Gemini, Azure OpenAI, Bedrock, Ollama, Copilot, Claude Code, Gemini CLI, Kilo CLI, and many OpenAI-compatible endpoints.

Keystore-Backed Secrets

Workspace-scoped provider API keys, MCP env/header secrets, superuser JWT signing material, and desktop gateway bearer tokens are stored in keystore.db, not in ordinary config tables.

Real Tools

Agents can use shell sessions, file reads and edits, patch application, grep, web search and fetch, URL downloads, computer use, MCP tool proxying, and dynamic skill tools.

MCP Servers

Install and manage servers compatible with Model Context Protocol per gateway and current workspace, track their health and catalog, and expose their tools to agents through the gateway. New servers are explicit-only by default and can be attached to a turn from the composer.

Skills

Pioneer supports skills compatible with the Agent Skills specification, including installation, validation, trust gates, dependency preflight, gateway and current workspace availability, upload flow, and health diagnostics. New skills are explicit-only by default and can be attached to a turn from the composer.

Tasks

Run scheduled and on-demand tasks with dependencies, retries, delivery state, progress events, write locks, task trees, and reviewable subagent result candidates.

Protocol-First Architecture

pioneer-protocol defines the public JSON-RPC surface and generated schemas under schemas/.

Explicit Artifacts

Files are workspace-scoped gateway artifacts. User uploads are copied to the gateway before a turn, and agent-created result files must be registered with the artifact tools before the final response. This keeps preview, download, and reuse workflows safe for both local and remote gateways.

Cross-Platform Packaging

Gateway builds target macOS, Linux, and Windows. Desktop packaging targets DMG, AppImage, and MSI.

Multi-Language Desktop

Desktop UI locales are available for English, German, Spanish, French, Hindi, Japanese, Russian, and Chinese.