Not a chat export. An append-only event log, derived facts with provenance, temporal claims, and queryable current state—built for multiple agents on the same project.
Graphe never treats raw chat as the system of record. Agents write events; workers derive observations and claims; retrieval serves current state with provenance.
A permanent log of what your agents did—tool calls, state changes, what context they used, and when things went wrong. Scoped to your team and project.
Facts extracted asynchronously from events—what happened, with provenance links back to source event IDs. Not rewritten when new events arrive.
Revisable beliefs about project state (confirmed, superseded, invalidated). Each claim carries lineage so agents and humans can see what evidence supports it.
The latest confirmed picture for agents to read before acting—not a chat transcript, not a single vector dump. Built from claims + graph + hybrid search.
Your agents send what they did—tool calls, errors, context used—into one reliable stream. Safe to retry; nothing silently dropped.
Ask what happened on a run, find similar past failures, and get explanations grounded in your project's actual history.
Browse runs, see what your agents believe about the project right now, and trace any claim back to the evidence.
Graphe turns raw activity into summaries, issues, and shared memory in the background—you don't manage pipelines.
From Python, TypeScript, Cursor, Claude, or your own setup. Technical setup guides live in Documentation.
Problem
Two agents overwrite each other's assumptions in the same repo.
How Graphe helps
Both write scoped events to the same project. Claims merge into one current-state view; supersession handles conflicts with provenance.
Problem
You need to know what context the agent had when it broke production.
How Graphe helps
Run timeline + retrieval snapshots on events + explain_run surfaces the failure chain without replaying the whole chat.
Problem
Agent acts on outdated file or claim state.
How Graphe helps
Hybrid search + current-state query returns confirmed claims and recent observations ranked by recency and relevance.
Sign up, connect your agents from the console setup guide, and ship your first instrumented run.