Context intelligence for software projects

Make Claude consume less context.While knowing more.

Turn what your project has learned into the context your next agent needs. Not another investigation from scratch.

Install OneCue See how it works
Your projectCode, decisions, discoveries
OneCue
Your coding agentsClaude Code now. More planned.

The code tells your agent what.
OneCue brings the why.

An old incident explains today’s retry guard. Without that connection, the next agent can repeat an investigation you already paid for.

From scattered evidence to useful contextIllustrative product walkthrough

What the project learned

Inspect the project evidence
decisions/payment-retries.md

A retry charged a customer twice.

The provider delivered the same event again. The handler treated it as a new payment.

Explains why the guard exists
Project understanding

A retry is not a new payment.

The constraint
Claim the event ID before applying payment. The incident explains why; the test defines what to preserve.
When it matters
Payment retries and webhook changes. Not unrelated work.

What reaches the next agent

Choose the next agent task

“Add retries when the payment provider times out.”

One relevant cue

Route retries through the event-ID guard before applying payment. A previous duplicate delivery charged twice.

The agent can build on the existing fix instead of repeating the incident.

Example sources and conceptual links. Automatic code-linked verification is planned; local capture and recall are the foundation.

A memory is useful only when it changes the next action.

See full session examples

The next layer Planned for Cloud

Not just remembered.
Kept up to date.

A saved lesson can become stale. We’re building background agents to verify, connect, and strengthen project understanding between coding sessions. The interface is your agent.

Explore OneCue Cloud
After a refactor Illustrative update

The guard moved. The rule still matters.

Previous locationsrc/payments/webhook.tsNew locationsrc/payments/claim-event.ts

Reconnect the incident and its test to the new handler, then update the next cue.

Verify Connect Refine
onecue doctor

Make context cost visible.

The first diagnostic we want to put in your hands: where Claude’s context budget goes, and what deserves a closer look.

Spend report planned. The doctor skill checks setup health today.

Claude Code Context ReportSample data
$96.20Example weekly spend

Potentially avoidable in this example

Repeated file reads
$21.40
Repeated repo discovery
$14.80
Duplicated architectural context
$9.10
Oversized CLAUDE.md / context
$7.60
Tasks suitable for cheaper models
$4.30
Illustrative opportunity$57.20 / 59%

Not measured savings or a forecast. Repeated work is not always avoidable; model pricing, caching, and OneCue’s own costs must be accounted for.

Give the next session
a better starting point.

Start with decisions, reasons, and relevant context for Claude Code. The free local skill is the foundation of the broader intelligence layer.

Start with the free local skill.

Install for Claude Code, then give it a decision worth remembering.

curl -fsSL https://onecue.sh/install.sh | sh
Free. Local files. No OneCue account.Full setup guide Skill README

Before you install.

What is OneCue?

OneCue is a context intelligence layer between your software project and the agents working on it. The local skill starts with decisions, reasons, and selective recall. The broader direction connects that knowledge to code and improves it through background verification and enrichment.

Is this just another memory tool?

Memory preserves information. OneCue’s goal is to determine why it matters, connect it to the project, keep it current, and select the context that should change the next action. The current local skill establishes capture and recall; code-linked verification and background enrichment are planned. CLAUDE.md and native memory can still complement it.

What works today?

One installer (curl -fsSL https://onecue.sh/install.sh | sh, or npm i -g onecue-cli) sets up everything in your repo: the skill, recall hooks, and the local store. The skill teaches the agent to capture and recall project memory as readable Markdown under .onecue/memories/. A companion doctor skill diagnoses the setup (npx skills add onecueai/onecue-doctor).

Does my code leave the machine?

OneCue’s local skill stores its records in your project, not in a OneCue cloud service. Claude Code still sends context to its configured model provider under your existing settings. OneCue Cloud is in development.

Which agents does it work with?

Claude Code and Devin today — `onecue install` writes hooks and the skill for both (.claude/ and .devin/); they share the same hook format. The skill also loads wherever Agent Skills are read. Codex, Cursor, Gemini CLI, and OpenCode support is planned through hooks, skills, and MCP integrations.

Can I get the doctor spend report today?

Partially. The onecue-doctor skill (npx skills add onecueai/onecue-doctor) checks setup health today — store, hooks, and runtime. The spend report shown here is a product concept with sample figures, not measured savings. Usage accounting and avoidability analysis are still planned.

What will OneCue Cloud add?

Shared project intelligence across machines and teammates, plus a background loop that checks saved lessons against changing code, connects evidence, and refines the cue your next agent receives.