Local context for AI work

detour.sh

When the context isn’t in the prompt, take a Detour.

Give your AI a safe, local way to retrieve the project rules, decisions, runbooks, prompts, and notes its harness doesn’t carry.

Stored locally. Retrieved through authenticated MCP. Shared only when you export.

Context with a destination

Stop rebuilding context for every session.

A Detour is a named set of context your AI can retrieve globally or for one project. Save it once, scope it clearly, and call it when the work needs it.

Save what matters

Keep rules, decisions, runbooks, prompts, and notes where your AI can retrieve them later.

Scope it correctly

Use Global Detours across your work and Project Detours only where their context applies.

Retrieve it safely

Connect a local AI tool through authenticated, read-only MCP without uploading your workspace.

Designed for AI workflows

The right context, retrieved on demand.

Your AI tool can ask Detour for relevant Global and Project context through a local, authenticated MCP connection.

In your Detours
Review before PRSkill
Project voiceSoul file
Release checksRunbook
Local MCP

Find and serve selected context

Your AI tool

“What should I know before this change?”

Relevant context, ready to use

Read-only retrieval with a key you create and control locally.

In your workflow

Bring context in before the work goes wrong.

Connect your local AI tool over MCP and retrieve the applicable Global and Project Detours before important work.

Returned material is context, not permission to execute commands or share data.

01Before action
02Before commit
03Before PR
04Before public

Under your control

Private at rest. Intentional when shared.

Your data stays in the container’s persistent volume. Select a context bundle only when you want to export it as Markdown or JSON.

Read the safety plan
Local by defaultDocker Desktop and a local SQLite volume
Selected exportsChoose exactly what goes into a bundle
No silent sharingSaving context never publishes it or sends it to an AI

Try the local build

Start on your own machine.

Save scoped context, retrieve it through MCP, and export selected bundles without a hosted account or external service.

docker compose up --buildGet the build