Rhei — Context infrastructure for coding agents

02 / Workspace

Every repo. One cortex.

Add product repos, dependency repos, old versions, examples. Rhei indexes each one locally and answers across all of them at once — while workspace memory compounds with every session.

›Where is checkout retry handled?6 indexes · 1 query

web

4.2k nodes

api

2.6k nodes

ui-kit

1.1k nodes

infra

640 nodes

legacy-v1

3.8k nodes

docs

320 nodes

◆2 repos · checkout/retry.ts + api/payments.ts · memory “retry boundary” reused

01

Cross-index search

One query fans out across every index in the workspace — product repos, dependencies, old versions — and comes back as a single ranked answer.

02

Memory that compounds

Every session feeds the evidence ledger. What agents reuse gets cheaper to find; what they contradict decays. The workspace gets smarter as it grows.

03

Quiet code stays cheap

New repos don’t bloat context. Cold code rests until evidence brings it forward, so ten indexes cost about as much attention as one.

03 / Agent Interface

Remember the work

Rhei is a self-tuning code graph memory for agents. Quiet code stays cheap, useful code becomes active when agents need it, and the graph learns from what agents actually use — keeping usefulness separate from truth, so memory earns its place instead of polluting context.

Why is auth failing across web + API?

Normal

████████████████████████████████████████

More tokens · still guessing

Rhei

█

1 tool call · paths, reuse, impact

Intent

Intent becomes a scoped goal with relevant files, components, constraints, and next steps.

Graph memory

The graph keeps quiet code cheap, brings useful code forward when agents need it, and learns from what agents actually use.

Control

Agents receive evidence before execution, while humans keep approval over what changes.

Trail

Goals, checkpoints, decisions, edits, and outcomes carry forward into the next session.

04 / Local

Local by default

Rhei starts on your machine. It maps your repo, builds local code intelligence, and exposes that context to coding agents over MCP. You can use local setup without an account. Pro features are optional. Hosted services for deeper context, team memory, and governance workflows only turn on after you opt in.

01

Local code map

Repo structure, symbols, relationships, and relevant context are built locally.

02

MCP for agents

Codex, Cursor, Claude Code, and other MCP clients can query your local code intelligence.

03

Pro is opt-in

Hosted services for deeper context, team memory, and governance workflows require sign-in and explicit approval.

05 / Enterprise

Trust by architecture, not promises.

Teams that cannot leak code do not need another assurance page. They need a system where the safe path is the default path — and every exception leaves a receipt.

local mode

Source stays local

In local mode, code, the graph index, reports, and memory never leave the machine. No account required, no source upload — by architecture, not by policy.

never · metadata_only · selected_slices

Explicit upload modes

Cloud features only see what the configured mode allows. “never” means never — features degrade to local behavior instead of leaking source.

default

No training on your code

Rhei does not train models on customer code by default, and non-zero-retention provider paths are labeled before they can be enabled for a workspace.

audit trail

Receipts, not assurances

Cloud assist, context decisions, and memory promotions are receipt-backed. You can see what was sent, what was used, and why — per request.

enterprise

Runs inside your walls

Private Docker / VPC deployment with SSO, access and policy controls, and audit support. The same local-first engine, governed at org level.

workspace memory

Governed memory

Team memory is scoped per workspace and member. Knowledge is promoted by corroboration, decays when contradicted, and stays inspectable.

Read the security overview Data Use contract Talk to us about a private deployment