Alfred Docs
Getting started
This guide gets Alfred running on macOS or Linux, then walks through the first local workspace loop: provider setup, source import, grounded Ask sessions, and automatic Dream maintenance.
The current public beta supports macOS Apple Silicon and Linux x86_64 with local workspace storage. Dream keeps routine maintenance automatic while logging risky or ambiguous actions as exceptions.
Requirements
Before installing, make sure the machine and trust boundary match the beta.
- macOS on Apple Silicon or Linux x86_64.
- A local workspace where Alfred can store settings, source files, and graph data.
- One of three inference paths: a subscription CLI (Codex, Claude Code), an Ollama install, or an OpenAI, Anthropic, or OpenRouter API key.
Install Alfred
Install from the signed release channel; macOS users can also use the Homebrew cask.
brew install --cask alfred-labs/alfred/alfred-app- Download the macOS DMG, Linux AppImage, or Debian package from GitHub Releases.
- Open Alfred after installation.
- Keep OS permission prompts visible so file access decisions stay explicit.
Open a workspace
Start with the Personal workspace. Alfred keeps imported files on disk and stores structured workspace records locally.
- Use Quick Note for a small first artifact.
- Use Dashboard to confirm the workspace shell loaded.
- Use Settings when you need to change appearance or provider choices.
Configure a provider
Open Settings → Inference. Alfred ships providers in one rail: subscription CLIs (Codex, Claude Code), the local Ollama daemon, and direct API keys for OpenAI, Anthropic, and OpenRouter. Provider choice is part of the workspace trust boundary.
- Subscription CLIs reuse the auth you already set up in your terminal (`codex login`, `claude login`). Alfred never reads or refreshes those tokens.
- Ollama connects to whatever local daemon you have running and uses any model you have pulled.
- API keys for OpenAI, Anthropic, and OpenRouter are stored in the OS secret vault (macOS Keychain) and used for pay-per-token calls.
Import sources
Use Inbox to add files while preserving the original source. Alfred extracts structure and evidence only after intake is traceable.
- Start with a few Markdown, PDF, JSON, YAML, CSV, or HTML files.
- Check imported items before relying on answers.
- Keep source files recoverable when testing extraction quality.
Ask with context
Use Ask to summarize imports, compare sources, search memory, or create notes from recent source material.
- Prefer questions that reference the current workspace.
- Inspect citations and retrieval traces before acting.
- Use the context panel to see sources, conversations, and session details.
Let Dream run automatically
Dream watches retrieval health, concept drift, knowledge gaps, hubs, and auditability in the background. Routine maintenance should feel automatic; exceptions are reserved for risky or ambiguous changes.
- Open Dream to read KPIs, trends, and the compact Autopilot log.
- Let routine cleanup and retrieval-health checks run without creating another inbox.
- Use exceptions, calibration, locks, and hubs only when Dream flags something that needs attention.
Features
Features
These guides mirror the public product docs: import sources, ask grounded questions, inspect relationships, and let agents reuse memory safely.
Ingestion
Bring files and notes into Alfred once, preserve the original source, and build searchable workspace knowledge from durable provenance.
- Markdown, PDFs, images, folders, quick notes, and MCP memory.
- Raw sources remain recoverable if extraction or indexing fails.
Ask
Ask turns local workspace evidence into answers with citations instead of treating every prompt as an empty chat.
- Blends FTS, lexical, vector, temporal, conversation, source, and document context.
- Best for incidents, product lookup, decisions, and cross-document comparison.
Graph
Graph maps sources, documents, concepts, hubs, and relationships so knowledge can be explored by connection, not only by filename.
- Shows imported sources, generated records, extracted concepts, derived hubs, and edges.
- Helps validate that imports are becoming structured knowledge.
Concepts
Concepts are evidence-backed ideas that connect documents using stable domain language.
- Useful concepts are short, specific, backed by sources, and reconfirmed over time.
- Hubs are the knowledge neighborhoods Dream derives from concepts and sources that recur together.
Dream
Dream is Alfred's quiet knowledge autopilot for retrieval health, concept cleanup, knowledge gaps, hubs, and auditability.
- Routine maintenance runs automatically or stays invisible.
- Exceptions are for risky or ambiguous changes, not the default workflow.
MCP
MCP exposes memory, search, document, source scope, and Dream tools so agents can use Alfred without opening the app UI.
- Agents can memorize context, recall prior work, inspect documents, and run Dream.
- Mutating tools are explicit and scoped.