Sibyl CLI
Sibyl is a self-hostable knowledge graph shared by Claude Code, Codex and Cursor for decisions, gotchas and conventions
- 68 ★ on the repository
- Last commit
- Apache-2.0
Extended access
Why this level
- Keeps project knowledge in a graph on your server that agents connect to
- Setup adds a skill and a SessionStart hook to Claude Code settings
- Ingesting mailboxes and archives brings their content into memory
Install
- Works in
Manual install
brew install hyperb1iss/tap/sibyl && sibyl setup https://sibyl.example.comFor macOS. Replace the server address with yours. The command signs in through the browser and installs the Sibyl skill and the SessionStart hook.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- Keeps project knowledge in a graph on your server that agents connect to
- Setup adds a skill and a SessionStart hook to Claude Code settings
- Ingesting mailboxes and archives brings their content into memory
This is third-party code. Review the repository files before installing.
What it does
Sibyl keeps knowledge about your projects in one graph that all coding agents use. Work goes through the shell: if an agent can run a command, it speaks Sibyl. The memory loop has four steps: sibyl context returns a compact pack with active work, decisions and constraints, the agent acts, records a finding with sibyl remember and then reflects. Semantic search finds knowledge by meaning. On top of the graph sit epic and task planning, document synthesis from your sources with citation and freshness checks, ingestion of docs and agent transcripts, and isolated namespaces for personal, project and team scopes. sibyl setup https://server-address signs in through the browser, installs the Sibyl skill for Claude Code, Codex and other agents and adds a SessionStart hook to Claude Code. If you have no server, the curl installer or brew install hyperb1iss/tap/sibyl and sibyl up start a local stack, and sibyl docker up deploys the API, web and SurrealDB. MCP is available, but the depth lives in the CLI.
Who it is for. Developers and small teams who work in several coding agents at once and do not want to retell the same context to each.
Good fit when
- You switch between Claude Code, Codex and Cursor, and each starts without knowledge of the project
- You want to record a bug's cause with sibyl remember and get it back through sibyl context on the next debug
- You want memory on your own hardware, not in a vendor's profile
- Epics and tasks have to live across sessions and teammates in one place
- Documents should be drafted from your own memory with citation and freshness checks
Not a fit when
- You want a light memory for one agent in a single file: Sibyl deploys a graph and a server
- Docker or SurrealDB cannot be installed: the web UI mode needs them, and a lighter option runs with --daemon
- Agents are not allowed to run shell commands: Sibyl's main interface is the CLI
- You want a fully managed SaaS memory with no administration: you run the server yourself
Example request
Connect this machine to our Sibyl server, record the expired Redis token bug in memory and show the context for debuggingLimitations
- Team use needs a server, which install.sh, sibyl up or sibyl docker up starts.
- Windows and other install paths are described separately in docs/guide/installation.md.
- sibyl setup signs in through the browser and adds a SessionStart hook to Claude Code.
- Ingesting sources such as mailboxes and archives widens the data that lands in the graph, so connect them deliberately.
How to disable. Stop the stack with sibyl down if you started it locally and remove the Sibyl skill and the SessionStart hook from the agents' settings. Data lives on your server and goes away with it.
FAQ
How is Sibyl different from agents' built-in memory?
A vendor agent's memory is tied to one tool and a vendor profile. Sibyl keeps the graph on your hardware and serves it to Claude Code, Codex, Cursor and your own agents. You can switch tools without losing what has accumulated.
Do I need MCP to use Sibyl?
No. The main interface is the shell: an agent that can run commands already works with Sibyl. MCP is there for clients that prefer it.
How do I install Sibyl on macOS and Linux?
On macOS run brew install hyperb1iss/tap/sibyl and sibyl setup with the server address. On Linux use uv tool install --upgrade sibyl-dev. With no server, start a local stack through install.sh or sibyl up.
What is the memory loop?
It is the order context, act, remember, reflect. The agent takes a context pack, acts, records a finding and then reflects, so the graph grows in every session.
What kinds of data does Sibyl store?
Decisions, gotchas, project conventions, epics and tasks, knowledge extracted from docs and agent transcripts. Data is separated into personal, project and team scopes.
Discussion
No questions or reviews yet.
Related
CLI toolsLong-term agent memory in one SQLite file with FTS5 search: an MCP server, a CLI and Hermes and LangGraph adapters
A memory and coordination layer for coding agents: specs and issues live in git while agents work in parallel
The Maximem Synap skill teaches Claude Code and Codex to add Synap long-term memory to your agent and wire it to the framework
Knowledge-graph memory for agents: Claude Code and Codex plugins, an MCP server and a Python library
Signet keeps a shared memory for AI agents: memories, session transcripts and secrets, with every fact linked to its source
Persistent agent memory in a folder of linked markdown notes, so Claude Code remembers a project across sessions
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| Sibyl (this card) | no | 68 | ||
| Sibyl Memory | no | API key | 177 | sibyl-memory is a separate project with a similar name, while hyperb1iss/sibyl has a graph with tasks and a CLI context, remember, reflect loop |
| Claude-Mem | no | 116,457 | claude-mem keeps one Claude Code's memory in a local database, while Sibyl shares it across agents and stores it on your server |
README in short
The README explains memory that works for its owner, then gives a capability table, a quick start for macOS and Linux, a local server, a first five minutes with remember and context and a description of the memory loop. Reference sections for the CLI, skills and MCP follow.