Local AI Research Tool

Build reusableAI
context.

Liner turns focused research into a durable Liner Project: a local working folder with a curated Mixtape inside. Define a Job to Be Done in Job Story form, Clarify Job, then add URLs, transcripts, PDFs, local files, or pasted text; Liner compiles the evidence and writes task-specific AI instructions.

V. 1.1.1Active

Instructions Written

Install

Open the builder,
in one command.

$npx linersh

Free · MIT · local · macOS, Linux & Windows

View on GitHub

/// v1.1.1

Sources selected /// task definedResearch compiled /// Markdown readyRunning locally /// no telemetryAI instructions /// LINER.md · SKILL.md · liner.yamlSources selected /// task definedResearch compiled /// Markdown readyRunning locally /// no telemetryAI instructions /// LINER.md · SKILL.md · liner.yaml

From ask to Project

One request becomes a reusable Liner Project.

Start with one narrow job. The builder helps define the evidence, approve the useful Sources, confirm the synthesis, and save both the research and the instructions a later AI session should follow.

Starting point

“Help my AI write App Store review-safe copy for a mental-health app.”

Clarify the job

Name the situation, desired outcome, and claims the AI should avoid.

Choose the evidence

Add review rules, health-claim examples, privacy guidance, and rejection notes.

Confirm what the AI inherits

After Sources are saved, you review the synthesis and boundaries before Compile. Liner records that accountable decision with the Project.

app-review-health-claims/

local files

liner.yaml
LINER.md
SKILL.md
mixtape/
  tape.yaml
  synthesis.md
  MIXTAPE.md
  sources/
    01-apple-review-guidelines.md
    02-rejection-postmortems.md
    03-medical-claims-boundary.md

MIXTAPE.md

Compiled research, synthesis, and links back to saved Source files.

LINER.md

Task-specific instructions for using the evidence and knowing where to stop.

SKILL.md

The Project entry point that lets a file-reading AI load the work directly.

Where it helps

Three jobs where source choice changes the output.

Use a Project when the task is narrow, the evidence needs judgment, and another session should be able to inspect the same foundation.

App developer

“Write App Store review-safe copy for a mental-health app.”

INPUT: Review rules, privacy guidance, rejected claims
OUTPUT: Evidence-backed rewrite instructions

Product designer

“Turn visual references into a concrete web art-direction brief.”

INPUT: Reference notes, layout patterns, web constraints
OUTPUT: A buildable direction with explicit boundaries

Engineer

“Plan a risky API migration from framework docs and past failures.”

INPUT: Official guides, breaking changes, postmortems
OUTPUT: A migration plan and review checklist

The path through the work

Four visible steps from task to Project.

Frame

Define the job in a plain sentence, then use Job Story and Clarify Job to sharpen it before research starts.

specific job

Gather

Bring in known material and review candidates. Before saving, Liner shows the Sources that will be added and leaves existing Sources unchanged.

source roles

Review

After the Sources are saved, confirm the synthesis before Compile. You can edit it first when it no longer represents the accepted evidence.

accountable judgment

Use

Add the AI instructions that travel with the research through Create Operating Layer.

Project Complete

Optional agent maintenance

Let Codex or Claude maintain the Project safely.

Install the Maintenance Adapter when you want a coding assistant to update an existing Project. The installed CLI remains the writer: it checks the saved Project identity, shows the exact change before approval, and records what happened after apply.

$liner adapters install codex --yes

Read the maintenance guide

Request

Add a Source, rename or move a Project, or archive a Project you no longer need.

Project Change Set

Review the exact effects before an approval-required change writes to the Project.

Change Receipt

Keep durable proof of the applied plan. Deleting a Project moves it to recoverable Liner Trash instead of erasing its files.

The right size of work

Use it when the evidence needs to outlive the chat.

Worth a Project

Source choice changes the quality of the output.
A later AI session needs the same evidence and boundaries.
The task is narrow enough to define and important enough to reuse.

A chat is enough

You only need one quick answer.
The evidence changes hour by hour.
The data belongs in a system outside a local folder.

Inspectable files

Sources, synthesis, instructions, and status stay visible as Markdown and YAML.

Local by default

The tool runs locally with no telemetry and uses your configured Codex or Claude account.

Portable with care

A .mixtape archive can include saved content, so check private Sources before sharing it.