Quickstart

Causal makes your agents smarter and more efficient the more you use them. It records what they learned, what they did, which outcomes followed, and what to change next time. It then serves that experience into the next decision.

It installs into the agent you already run:

Claude CodeCursorCodexOpenClawHermesGitHub CopilotGemini CLIOpenCode

All eight take the memory skills. The first five also get MCP wiring and the always-on rule today; Copilot, Gemini CLI, and OpenCode get those as each one’s config support settles.

Three steps to get there.

1. Install it where you already work

One command, from the project you want your agent to remember.

npx causaldotstore

It signs you in, detects the agent runtimes you have, and wires in the rule that recalls experience before your agent acts. What it prints when it finishes:

Installed 2 skill(s) into 3 runtime(s):
  Claude Code      → ~/.claude/skills  (2 skills)
                   MCP → ~/.claude.json  (wired)
                   rule → ~/.claude/CLAUDE.md  (appended)

Eight runtimes are supported — Claude Code, Cursor, Codex, GitHub Copilot, Gemini CLI, OpenCode, OpenClaw, and Hermes — and anything outside them uses the TypeScript or Python SDK. Every path the installer writes is on Install.

2. Keep working

Give your agent the same tasks in the same harness. Causal records as it goes: the decision, the action that followed, and the outcome it produced. Your part is to work. The graph fills behind you.

3. Let it compound

The first sessions build the record. After that, recall starts paying. A task that resembles earlier work opens with the relevant precedent already in context, and an approach that failed last week comes back labelled with what it cost. Every session adds to it, so the agent you hand work to next month starts from more than the one you started with today.

Get more out of it

Add a second agent. Point Cursor at the same graph Claude Code has been filling, and it starts from what that work already established: Install.
Bring in the record you already have. Connectors read Slack, GitHub, Jira, Linear, Notion, Salesforce, and SAP, so the graph compounds from work done before your agents ran once: Integrations.
Check what came back. Every served precedent traces to the work that produced it: Recall and capture.