Quick Start

This guide takes you from nothing to a running Sibyl with your first captured memory, in about five minutes.

Working on Sibyl itself? This guide is for using Sibyl. To set up the monorepo for

development, see Installation. :::

Step 1: Install Sibyl

The shell installer starts the local API + web stack and opens the setup UI:

curl -fsSL https://raw.githubusercontent.com/hyperb1iss/sibyl/main/install.sh | sh

Already use Homebrew? Install the package, then start the local UI:

brew install hyperb1iss/tap/sibyl
sibyl up

Step 2: Open Sibyl

Service	URL
Web UI	http://localhost:3337
API + MCP	http://localhost:3334

Step 3: Finish setup in the browser

The first time you open the web UI, a setup wizard runs. It walks you through three things:

1. API keys: Sibyl needs an Anthropic key for entity extraction and an OpenAI or Gemini key for embeddings.
2. Admin account: the first account, which holds owner privileges.
3. Connect: how to start using Sibyl from the terminal or an agent.

Step 4: Connect the CLI

Confirm the CLI can reach your local server:

sibyl health

Sibyl is now yours from any terminal.

Step 5: Run the memory loop

Sibyl's core is a loop: recall, act, remember, reflect. Try it.

Capture something worth keeping:

sibyl remember "Async gotcha" \
  "Use asyncio.gather for concurrent awaits, not a sequential loop" \
  --kind pattern

Pull it back as working context:

sibyl context "async concurrency" --intent build

Or load context across every accessible project:

sibyl context "review prior async concurrency lessons" --intent review --all

Semantic search finds that memory even though you searched with different words.

Pull the full record back with sibyl show <id>. You can also seed memory from past agent sessions: sibyl ingest claude-code <path> and sibyl ingest codex <path> import transcript JSONL.

Step 6: Connect your AI agent

Sibyl earns its keep when your AI agent uses it too. Any agent can reach Sibyl through the sibyl CLI, and MCP-capable agents (Claude Code, Codex, opencode, OpenClaw, and others) can connect to the MCP endpoint.

Use the Connect page in the web UI for per-client MCP configuration and the agent prompt snippet, or see Agents & MCP.

Where to go next

• The Memory Loop: recall, act, remember, reflect
• Capturing Knowledge: what is worth saving
• Task Management: plan and track work
• Agents & MCP: connect any AI agent

Common commands

Action	Command
Capture a memory	sibyl remember "Title" "What matters"
Load context	sibyl context "goal" --intent build
Search broadly	sibyl context "query" --all
Show a record	sibyl show <id>
Create a task	sibyl task create --title "..."
Complete a task	sibyl task complete <id> --learnings "..."
Link a repo	sibyl project link <id>
Check health	sibyl health