Future Equals NowLabsContext Atlas

A HOME FOR EVERY INSTRUCTION

Context
Atlas

Put the right instructions in the right place.

Permanent rules, project state and deep reference material often end up in the same file. Give each one a clearer home.

LESS TANGLED. MORE INTENTIONAL.

AGENTS.mdone starting point
Root rulesalways relevantProject statewhat’s currentDeep knowledgewhen neededWorkflowsrepeatable tasks

Keep the rules. Make the rest discoverable.

Why organise agent context?

Your agent may need different information at different times. Keep durable rules visible, move changing project state out of permanent instructions, and make deeper knowledge and repeatable workflows discoverable when needed.

An instruction can be important without belonging in every task’s persistent context.

Private by design

Local browser analysis · No cloud AI analysis · Nothing uploaded

Your files remain in this tab’s memory. No account, source-content telemetry or stored session. Refresh or reset clears your inputs.

How this works

Bring your agent context.

Drop in your docs or paste instructions. We’ll map the structure locally.

Drop .md or .txt files here

AGENTS.md · CLAUDE.md · Skills · supporting docs

0 / 80,000
Pasted context role · optional
Formats & limits

UTF-8 .md / .txt. Up to 12 inputs, 80,000 characters per input, 240,000 characters and 8,000 lines total. Files: 160 KB each, 480 KB per selection. Paste and files can be analysed together. Designed primarily for English-language instructions. File roles are inferred from names; you can adjust them after selecting files.

Private by design · No cloud AI analysis · Nothing uploaded

Private by design. How this works.

Local browser analysis · No cloud AI analysis · Nothing uploaded. Your source stays in this tab’s memory. A local Web Worker runs deterministic checks; if it’s unavailable, the same engine runs in this tab and yields between stages. No accounts, telemetry or stored sessions. Refresh or reset clears your inputs. Copy and download create copies only when you choose. No LLM or external model call is used.

Patterns describe supplied text and suggest possible homes. They cannot establish correctness, runtime loading or repository completeness. Keep uncertain constraints and review all moves.

Why organise context instead of just shortening it?

Think of context as layers an agent may need at different moments. Put durable rules where they remain visible, and make deeper information discoverable for the tasks that need it. These are suggested roles, not a promise about any agent’s loading behaviour.

Universal rules
AGENTS.md / CLAUDE.md
Current truth
PROJECT_STATE.md
Deep knowledge
docs/
Repeatable procedures
skills/
Subsystem rules
Scoped instructions
One-off requirements
Task prompt

An instruction can be important without belonging in every task’s persistent context.

Reorganise knowledge, retain safeguards, and check the result on real tasks.