What it's for

Use cases

Each one is the same engine — memory, live data, and the graph that links them — doing real work for a coding agent. They build along the spine: memory first, then the live data it's grounded in, then the graph that connects it all, then anything else your org runs on.

Memory

Your agent remembers across sessions

Stop re-explaining your codebase. Architectural decisions, conventions, and gotchas persist and get recalled into every prompt — across sessions and machines.

pensieve remember "Auth is Supabase JWT; prefer KQL over SQL in examples."

Onboard an engineer's agent in a day

Point a new hire's agent at the team's shared memory and the repo graph, sync once, and it already knows the architecture, the services, and the conventions.

PENSIEVE_CLOUD_URL=https://pensieve.your-co.dev pensieve sync

Live data

Debug a prod incident from your editor

The agent recalls similar past incidents, queries live logs and traces, and walks the service graph to the failing call — without you leaving the IDE.

otel_logs | where _timestamp > ago(15m) and severity_text == "ERROR"
| summarize n = count() by service_name | order by n desc

Ask your whole stack in plain English

"Why is checkout slow today?" → the agent turns it into KQL/SQL across logs, traces, data sources, and your operational databases, and streams the answer.

pensieve query "p99 latency on /api/checkout in the last hour, by version"

The graph

Trace one customer across every service

From a customer to their auth events to the deploy that broke their checkout — one graph traversal, not five dashboards open side by side.

context_edges | graph-traverse source "cus_42" from src to dst max-hops 3

Turn your repo into a queryable graph

Connect GitHub and the agent traverses code, PRs, issues, and people — and your memories link to the real files and services they're about.

pensieve datasource add github your-org/your-repo --start

Anything

Make any source queryable

Webhooks, billing exports, CI outcomes, your own metrics — one POST and it's queryable, agent-readable data that prunes well from day one.

curl -X POST $PENSIEVE/v1/ingest -H "X-Table: github_actions" --data-binary @runs.ndjson

Build your own

These are starting points, not a closed list — anything with a timestamp and a body becomes memory-adjacent, queryable data. The mental model is How pensieve works; the surfaces are Ingest, Query, and Connect your agent.