Agent memory with integrations
Causal is experience memory for AI agents — it records what your agents tried, what happened, and why, then serves that experience into the next decision. Integrations bring in experience from outside your agents' own work: seven business systems the company already runs, read for the outcomes they hold.
The seven systems
Slack, GitHub, Jira, Linear, Notion, Salesforce, and SAP.
Causal does not compete on connector count. Each of the seven is there because the company's record of what it tried and what it cost already sits inside it — the merged-then-reverted pull request, the reopened ticket, the postmortem, the decision that later shipped. The graph starts compounding from work already done, before your agents have run once.
Ingest is episodic and firewalled: nothing a connector reads can become an instruction, so a hostile message in a channel cannot program your agents. How a poisoned write is adjudicated after that is on Security and memory integrity.
What each system contributes
Each system holds a different part of the same record. The examples below trace one checkout-latency incident across all seven. They are illustrations, not real records.
| System | What it contributes | Illustration |
|---|---|---|
| Slack | The channel where a decision was made and its result came back | An incident thread where the team chose to raise cache limits and checkout latency spiked again |
| GitHub | Code changes and what became of them | A pull request merged and then reverted, because the cache keys assumed the old config |
| Jira | The ticket lifecycle, including the tickets that do not stay closed | An incident closed after the hotfix and reopened the next day on a stale flag |
| Linear | Cycle and retro records: what the work displaced | A cycle retro where unplanned incident work pushed auth two weeks late |
| Notion | Written analysis after the fact | A postmortem tracing the whole chain back to a cron job that overlapped the deploy window |
| Salesforce | The account side of an outcome | A renewal that stalled because two P1 escalations sat unresolved |
| SAP | ERP order and stock actions | A stock hold placed on a product variant in one warehouse, and procurement for a cache vendor that stalled nine days on missing compliance forms |
Read in order, those seven entries are one history rather than seven feeds. Each carries the outcome of an action and the change that followed, so a revert stays attached to the merge that caused it and a reopen to the fix that missed. An agent asked why a stock hold exists can point at the incident.
The shape of an ingested memory
A connector does not dump a transcript. It produces the same shape the rest of the graph uses — cause, effect, and the experience derived from them — carrying four fields: the source it came from, the event that occurred, because (the cause), and next (what should change).
As an illustration, a GitHub connector reading a reverted pull request would yield a memory along these lines:
| Field | Value |
|---|---|
| source | PR #4821 · reverted |
| event | Rushed the policy fix; it broke the release pipeline. |
| because | Cache keys assumed the old config. |
| next | Canary config changes before rollout. |
That example is written to show the shape. It is not a real record.
Because each memory carries its cause and its consequence, precedent comes back attached to the outcome it caused. Causal serves precedent and evidence. Your agent still reasons about what to do.
How an ingested memory reaches the next decision is on Recall and capture.
What is not shipped yet?
None of the seven is shipped. Connectors land with the trial.
Connector authorization, scopes, polling or webhook behavior, and per-connector configuration are not yet published. This page will carry them when they are.
Causal installs into the harness you already run — Claude Code, Cursor, Codex, OpenClaw, or Hermes — and Python and TypeScript SDKs cover anything outside that list. Both paths are on Install. The SDKs, the MCP server, the node and export format, and causal-verify are open source; the engine that turns raw agent activity into adjudicated experience is not.