A.D.A.M. is a behavioral specification you load into any AI chat host that supports a persistent instruction field. Once active, every reply is routed through one of three explicit depth modes: LOW for fast minimal answers, MID for structured default responses, and DEEP for full decision support. Depth is chosen by analyzing the shape of your message — not its keywords — and you can always override it manually.Documentation Index
Fetch the complete documentation index at: https://mintlify.com/XxYouDeaDPunKxX/A.D.A.M.-Adaptive-Depth-and-Mode/llms.txt
Use this file to discover all available pages before exploring further.
Quickstart
Activate A.D.A.M. in 30 seconds and send your first depth-routed message.
How It Works
Learn how structural routing, mode selection, and the output contract fit together.
Core Concepts
Understand LOW, MID, and DEEP modes, the state machine, and routing rules.
Command Reference
Every command, probe, and manual override — with exact syntax and output contracts.
What A.D.A.M. changes
Without a depth protocol, a chat session can drift: responses grow verbose, corrections pile up, and the assistant expands every topic by default. A.D.A.M. solves this by making depth explicit and structural.⚡ LOW
Fast, minimal answers for factoids, definitions, and banter. No follow-up questions unless they materially change the answer.
🧭 MID
Default structured mode for explanations, comparisons, and moderate evaluations. Compact structure only when it improves clarity.
🧠 DEEP
Full decision support: explicit assumptions, comparisons, failure modes, stress tests, and a bounded AUDIT footer.
Get started in 3 steps
Download the protocol file
Open
A.D.A.M.v5.txt from the repository. This is the canonical spec — the entire protocol is contained in this single file.Load it into your chat host
Paste the full contents of
A.D.A.M.v5.txt into your host’s persistent instruction field (system prompt, custom instructions, or equivalent). The file must be fully visible in context — truncation will cause fail-closed behavior.A.D.A.M. is not a library or a service — it is a portable text specification. There is nothing to install and no API to call. The spec is the implementation.
Why it works
A.D.A.M. routes by message structure, not semantics. It counts option blocks, step sequences, criteria items, and numeric constraints to determine whether a request warrants deeper reasoning. This means simple requests stay light automatically, and complex requests trigger a consent-gated DEEP pass rather than silently expanding.Deep Gating Guide
How POSSIBLE DEEP detection and the consent gate work in practice.
Structural Routing
The structural kernel primitives that drive automatic mode selection.
Recovery and Remount
How to recover an A.D.A.M. session after context loss using ADAM REMOUNT.
Host Compatibility
Known constraints and what to check before deploying A.D.A.M. on a new host.