Documentation

Use, connect, and understand Zharwing Memory.

Setup, daily use, agent integration, architecture, and security — all of it here.

01Set up Memory

Choose a local profile and start the interface you need.

Open setup →02Connect an AI agent

Point Codex, Claude, or another MCP client at your local Memory.

Open MCP setup →03Review decisions

See why the system is built the way it is.

Open decisions →

All guides

Browse by task.

Start here

Project overviewWhat Memory does, how it is organized, and the fastest way to start.SetupChoose a supported local profile and start the daemon, browser UI, desktop app, or agent connection.Frontend V2 statusSeparate implemented source boundaries from candidate and release/device qualification.Developer previewWhat is ready today, current limitations, and the project boundary.First walkthroughCreate a project, start a session, save context, and resume work.Browser UIRun the frequently used local interface in a browser.Desktop UIUse the native application, navigation, and folder selection workflows.

Use Memory

User flowsUnderstand setup, daily sessions, context, review, and recovery flows.RepositoriesConnect one or several code repositories to a Memory project.WorkstreamsGroup sessions and knowledge around long-running areas of work.Import existing memoryPreview and import existing Markdown documents and session history.Graph rulesShape useful graph hubs with deterministic project rules.Semantic graphAdd locally reviewed semantic relationships to the context graph.

Connect AI agents

MCP setupConnect Codex, Claude, or another MCP-compatible client.Agent automationBootstrap repositories and make session memory part of agent work.Agent protocolFollow the shared startup, checkpoint, and closeout contract.Generic MCP adapterUse the project-neutral MCP integration pattern.CLI adapterUse Memory through command-line workflows.JSON-RPC adapterIntegrate directly with the local daemon API.

Architecture and decisions

ArchitectureSee the local daemon, adapters, packages, and trust boundaries.Data modelUnderstand projects, sessions, documents, graph data, and storage.API referenceReference the local HTTP, JSON-RPC, and MCP surfaces.System diagramsReview the main component, data, and interaction diagrams.Source and context boundaryUnderstand which sources become public documentation and how private context stays excluded.Architecture decisionsRead the decision log and the rules for changing durable system boundaries.ADR 0001: Local-first project boundaryKeep application source, private memory, and public documentation in separate ownership domains.ADR 0002: Progressive public documentationGenerate directly addressable pages whose content and navigation do not depend on JavaScript.ADR 0003: Project-bound agent authorityUse explicit project-bound agent authority and centralized privacy projection.

Security and contracts

SecurityUnderstand the local trust model and report vulnerabilities.Principal modelReview audiences, project binding, expiry, revocation, and operation authority.Browser session securityReview local browser bootstrap, CSRF, rotation, revocation, and project binding.Native desktop authorityReview the Rust-owned daemon and credential boundary outside the webview.Provider secrets and egressOperate write-only credentials and constrained provider destinations.Privacy projectionUnderstand centralized agent and provider projection, visibility, and completeness.Operation admissionTrace requests through exact decoding, authority, idempotency, and dispatch.

Operate and contribute

OperationsRun, package, back up, restore, and troubleshoot Memory.AI provider testingTest local and external model-backed workflows safely.Testing planSee the testing strategy and expected coverage.Frontend accessibilityReview the responsive, keyboard, focus, contrast, motion, and manual qualification contract.Frontend V2 migrationMigrate profiles, credentials, visibility, preferences, and layout caches with rollback.Compatibility registerSee retained and removed compatibility paths, callers, dates, and removal proof.Frontend qualificationSeparate source observations from candidate, browser, WebView, assistive-device, and release evidence.ContributingSet up the repository and follow the contribution workflow.