Engram.AI CLAUDE.md

Engram.AI CLAUDE.md is an instructions file for coding agents from DPBG/Engram.AI. It costs 2,346 tokens per session, scanned A, original, MIT.

Repository instructions for Engram.AI, a continuously learning AI system made of a neural-network core and supporting services for memory, planning, safety, and a web dashboard.

In plain words
What is it for?
Use them when modifying Engram's Python services, neural-network code, NATS communication, databases, Docker setup, or dashboard.
Why use it?
They explain the project's non-negotiable architecture and tell coding agents when to stop before making a conflicting change.

Instructions file

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

agentmods
npx agentmods add instructions/dpbg/engram.ai/claude-md
Clone the repo
git clone --depth 1 https://github.com/DPBG/Engram.AI

Wrote this? Show the measurements

A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.

agentmods badge for Engram.AI CLAUDE.md

README.md
[![agentmods](https://agentmods.dev/badge/instructions/dpbg/engram.ai/claude-md.svg)](https://agentmods.dev/instructions/dpbg/engram.ai/claude-md)
Your own site
<a href="https://agentmods.dev/instructions/dpbg/engram.ai/claude-md"><img src="https://agentmods.dev/badge/instructions/dpbg/engram.ai/claude-md.svg" alt="Measured on agentmods" height="20"></a>
Per session 2,346 This file is loaded in full into every session.
When invoked 2,346 The same file — it is already loaded in full.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5 $0.02346 $0.02346
Opus 5 $0.01173 $0.01173
Sonnet 5 $0.00469 $0.00469
Haiku 4.5 $0.00235 $0.00235

Measured 4d ago against content hash 1ffbef2678c8, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

Engram.AI CLAUDE.md scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 4d ago.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

CLAUDE.md · 191 lines

How it starts

The opening of the file, as written. The whole thing — 191 lines — stays where its author put it; the contents beside it link to each section on GitHub.

CLAUDE.md — Engram Architecture & Contributor Guide

This file is the authoritative source for Engram's non-negotiable architectural constraints. It is cited by CONTRIBUTING.md, DESIGN-PRINCIPLES.md, and the issue templates. It also serves as the working context for Claude Code and other agents operating in this repo.

If a proposed change conflicts with anything here, stop and open an issue before implementing.


1. What Engram Is

Engram is a self-aware, continuously-learning neuromorphic AI system. Its intelligence lives in a spiking neural network (the neuromorphic/ "brain"), surrounded by microservices that provide sensory input, safety governance, memory, planning, and a web dashboard. Services are independent processes that communicate over NATS and persist to SQLite (+ Qdrant for vectors).

Two ways to run the same system:

  • Pure Python (python run.py) — the launcher downloads NATS and runs each service as a subprocess. Best for local development. See RUN-LOCAL.md.
  • Docker Compose (docker compose up) — each service is a container. The Hetzner deployment layers docker-compose.yml + deploy/docker-compose.1m.yml.

2. The Six Architectural Invariants (non-negotiable)

All neuromorphic code MUST conform to these. They define what Engram is; a change that violates one is a change to a different system. Primary implementation files are listed for each.

Invariant 1 — Integrated Multi-Mechanism Learning

All 6 learning mechanisms operate together and are never individually disabled: STDP, eligibility traces, BCM metaplasticity, 4-channel neuromodulation (DA/ACh/NE/5-HT), homeostatic scaling, and R-STDP.

  • Files: neuromorphic/src/neuromorphic/synapses.py, neuromodulation.py, network.py
  • Enforcement note: For performance, some mechanisms update on a fixed interval (e.g. STDP every N steps) using compensated decay so the result is mathematically equivalent to running every step. This is "logically every step." Any change to those intervals MUST preserve equivalence and be covered by an equivalence test. Mechanisms may never be turned off.

Read the full file on GitHub · 191 lines

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 4d ago First seen · 191 lines · 2,346 tokens per session scan A 1ffbef2678c8

Subscribe to this mod's changes

Engram.AI CLAUDE.md is an instructions file published in the GitHub repository DPBG/Engram.AI (5 stars, last pushed 1mo ago), licensed MIT. It adds 2,346 tokens to every session, about $0.0117 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.