Brain.db
Persistent memory. SQLite, FTS5, vector embeddings, 3-signal RRF recall, ACT-R decay, graph edges.
typescript · sqlite
KERNL
Session persistence, semantic memory, cross-project learning, adversarial testing, crash recovery.
typescript · 128 tools
CONTINUITY
Crash recovery, decision registry, context compression for AI sessions.
typescript · sqlite
oktyv
Browser, API, shell, email, database, cron, vault, visual QA, parallel execution via MCP.
typescript · 10 engines · 73 tools
SHIM
Code quality analysis, multi-AI coordination, distributed systems architecture.
typescript · redis · bullmq
AEGIS
CPU, IO, memory, power optimization for AI-heavy workflows. Five canonical profiles.
typescript · windows
NIGHTSHIFT
13-pass autonomous memory maintenance. ACT-R decay, anti-Hebbian pruning, structural isomorphism, epistemic scanning.
node.js · autonomous · 13 passes
AUTONOMIC
Autonomous sprint execution. Protocol, CLI, and plugin for AI agents. 4-gate quality system, tiered execution, structured abort contract, learning loop.
node.js · protocol · cli
Surface Tag
Browser extension that auto-tags Claude messages with device/browser metadata for surface-aware conversations.
javascript · chrome extension
Six of these compose in sequence — a query is routed, deliberated, checked for premature convergence, gated, and scored before anything ships. homeostasis sets the operating mode the others work under; DAP handles cross-model verification, outside the single-query chain. Each is documented and versioned independently below.
query → oracle-router → tribunal → signal-diversity → veto-authority → composite-confidence-score → assertion-router → output
SafeWord
Credential manager for logins and API keys alike — same discipline for both, not one careful and one casual.
credential management · in development
YUMA
Testing and validation infrastructure, named after Yuma Proving Ground — the idea being: if it survives Yuma, it survives anything. A pattern, not a single codebase, built into the tooling wherever code quality gets checked. Health scoring combines feature completion with test coverage and recency, so a project can't read as "done" on features alone while its tests have quietly gone stale — and commits get gated on the result rather than trusting that tests were run.
pattern · implemented across several systems
Morpheme-G
AI content generation engineered for genuine variation at scale — structure and voice differ in real ways, not just synonym-swapped surface text.
content generation · diversity engineering
ROSETTA
An ambient-context daemon with a specific privacy architecture built in from the start, not bolted on after: raw source content is processed locally and never leaves the local network — only the resulting distilled observations sync outward. Privacy escalates up from the default, never down. The interesting part isn't that it ingests context; it's that the boundary between what stays local and what's allowed to leave is enforced structurally, not by policy someone has to remember to follow.
ambient intelligence · privacy architecture
Assertion Router
4-mode confidence routing. ASSERT, PROBABILISTIC, INVESTIGATIVE, REFUSE based on epistemic certainty.
typescript · decision theory
Composite Confidence Score
7-axis weighted confidence function. Evidence, reasoning, calibration, source, domain, coherence, meta.
typescript · equation
Tribunal
Multi-perspective deliberation. Structured adversarial roles: prosecutor, defense, journalist, academic.
typescript · adversarial reasoning
Homeostasis
Operational mode state machine. CRISIS, DEEP_WORK, EXPLORATION, CONSOLIDATION, RECOVERY.
typescript · state machine
ORACLE Router
Entropy-minimizing action dispatch. Routes to the action that reduces the most uncertainty per unit cost.
typescript · information theory
Veto Authority
Quality-floor veto with temporal decay R(t). Prevents low-confidence outputs from propagating downstream.
typescript · quality gate
Signal Diversity
Novelty injection and outlier preservation. Prevents premature convergence in reasoning systems.
typescript · divergence
DAP
Deliberative Assembly Protocol. Verification residuals for multi-model AI. Prediction errors propagate, agreements explain away.
specification · protocol
Five specialized engines behind a research and mapping platform — automated enrichment, periodic re-scanning, cross-source confidence, media-attention tracking, and anomaly detection, each doing one job rather than one system trying to do all five.
MERIDIAN
Automated research enrichment. Pulls continuously from open government and public-record data sources to build context around tracked subjects, rather than waiting for a human to go looking.
pattern · part of a research platform
SENTINEL
Periodic re-scanning of tracked subjects for meaningful change. Most research is a snapshot; this treats it as something that needs to be checked again, on a schedule, without anyone having to remember.
pattern · part of a research platform
PARALLAX
Confidence from triangulation, not a single score. Weighs a finding by how many genuinely independent angles it's visible from, the way a fact confirmed by unrelated sources carries more weight than one repeated by a single one.
pattern · part of a research platform
Sentry
Ongoing media-attention tracking for tracked subjects — not just what's being said, but how the volume and pattern of coverage shifts over time.
pattern · part of a research platform
ATHENA
Statistical anomaly detection across a large evidence set — surfacing the pattern that's genuinely unusual against its own baseline, which is easy to miss by reading case by case.
pattern · part of a research platform
EoS
AST-level code quality scanning, structural analysis, dependency mapping, health scoring.
javascript · ast
SCRVNR
Voice synthesis. Learns authentic writing style, calibrated separately across different registers of a person's writing.
python · nlp
VIGIL
Video intelligence. Download, transcribe, scene detection, keyframe capture, frame analysis.
python · whisper · ffmpeg
dedact
Forensic PDF analysis. De-redaction detection, document extraction, metadata analysis, entity resolution.
python · ocr · spacy
WHETSTONE
Adversarial modeling as a design principle, not a single codebase. Before a conclusion ships, something constructs the strongest actual case against it — not rhetorical devil's advocacy, a genuine counter-position evaluated on its own terms. Implemented separately wherever it's needed rather than centralized in one place.
pattern · implemented across several systems
TESSRYX
Universal dependency intelligence. Provenance tracking and blast-radius analysis — traces what a conclusion actually derived from, and what breaks downstream when something upstream changes.
python · dependency graph
IMPRINT
Structured post-session reflection as a recurring pattern, not a single codebase — what changed, why, and what should change next, captured as typed deltas rather than left to live only in conversation history.
pattern · implemented across several systems
CORTEX
A unified inference and memory layer that routes by sensitivity before it routes by task — the more private the content, the more locally it stays, escalating to broader models only when the question actually calls for it. Forms durable beliefs rather than just logging events, and generates its own questions when something doesn't add up, rather than waiting to be asked.
inference routing · memory · not yet a standalone repo
cognitive-stack
Architecture specification for the full 10-system persistent AI cognitive organism.
documentation · 10 systems
HIRM
Mathematical consciousness framework. Synthesizes IIT, GNWT, FEP, SOC, and HOT into a unified formal model.
python · formal model
Multiplicative Composition
Axiomatic proof that independent dimensions compose multiplicatively in emergent systems. Eight domains tested.
python · information geometry