A.D.A.M. (Adaptive Depth and Mode) is a control protocol you load into any AI chat host that supports a persistent instruction field. Once active, every reply is tagged with a visible mode —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.
LOW, MID, or DEEP — and the assistant’s depth of reasoning is driven by the structure of your request, not by its own defaults.
The result: less noise, clearer answers, and explicit control over when to invoke full decision support.
Introduction
Understand what A.D.A.M. is, why it exists, and how it changes AI chat behavior.
Quickstart
Load the protocol and send your first
ADAM PING in under a minute.Core Concepts
Learn about the three depth modes, structural routing, and the output contract.
Command Reference
Every public command and manual override, with exact syntax and expected output.
How it works
Load the protocol
Open
A.D.A.M.v5.txt and paste its full contents into your AI host’s system prompt or instruction field.Activate with ADAM PING
Send exactly
ADAM PING. The assistant replies with a 4-line bootstrap output including ADAM_PING_OK and a BOOTSTRAP_CLASS indicating how the protocol was loaded.Ask normally — depth is automatic
Ask any question. Simple factoids get a
MODE: LOW reply. Structured requests with options, criteria, or numeric constraints trigger MODE: MID or route toward DEEP. The mode tag appears on the first line of every reply.What A.D.A.M. changes
Without A.D.A.M., AI chat tends toward over-explanation: every question gets a long answer, every answer includes caveats, and sessions drift over time. A.D.A.M. imposes a visible, rule-ordered contract on how the assistant responds.Visible depth
Every reply opens with
MODE: LOW, MODE: MID, or MODE: DEEP — you always know what kind of answer you’re getting.Structural routing
Routing is driven by message shape (options, steps, criteria, numbers) — not keywords or tone.
Consent-gated DEEP
Full decision-support mode only activates after you explicitly approve. No accidental escalation.
Bounded audit
Complex replies include a strict
AUDIT footer with ACTION, RISK, and BASIS — or a dash when grounding is absent.Fail-closed
When strict invariants can’t be satisfied, the protocol outputs
ADAM_UNSUPPORTED instead of silently degrading.Session recovery
Use
ADAM REMOUNT to generate a recovery capsule. Paste it back at any time to restore the ACTIVE state.Key sections
Using A.D.A.M.
Commands, overrides, DEEP gating, and session recovery — everything you need to operate the protocol day-to-day.
Diagnostics
Self-test, system status, input tracing, and failure diagnosis tools for verifying protocol health.
Reference
Official terminology, bootstrap class definitions, host compatibility notes, and a real-world field report.
Output Contract
The exact rules governing mode tags, AUDIT blocks, provenance signals, and fail-closed behavior.