Hyperlex

Catch language while it’s still becoming culture.
Hermes skill · memetic emergence engine · settled forecasts only

Home · Quickstart · See it work · Commands · Operator loop · Run history · Status

_{Offline first success (no API keys): python3 scripts/hyperlex.py demo}

_{Visual · Grok Imagine}

Plain English

Hyperlex is a cultural radar for slang.

New phrases don’t show up fully formed. They leak out of group chats, betting forums, crypto timelines, and AI-native meme stacks — half joke, half signal. Hyperlex watches that fog and asks practical questions:

What just showed up?
Which family of slang does it belong to?
How hard is it spreading?
Is it just noise, or is it starting to make itself real (hyperstition)?
If we forecast, can we score ourselves later — honestly — when the world answers?

It runs as a Hermes skill: offline-friendly by default, receipt-backed when you care, and stubborn about one rule — no fake accuracy. Brier scores only appear after a human settles an outcome. Simulation stays labeled research. History doesn’t get rewritten.

Cool without the con: less “AI vibe report,” more flight recorder for memetic weather.

Atlas

How the pieces sit together:

                    ┌─────────────────────────────────────┐
                    │           HERMES SKILL               │
                    │   SKILL.md · scripts/hyperlex.py     │
                    └─────────────────┬───────────────────┘
                                      │
     ┌────────────────────────────────┼────────────────────────────────┐
     ▼                                ▼                                ▼
┌──────────┐                   ┌────────────┐                   ┌────────────┐
│  INTAKE  │  --route offline  │  ANALYSIS  │                   │  RESEARCH  │
│  routes  │  --route live ──► │  lineage   │                   │  Phase 5   │
│  mock ·  │                   │  virality  │                   │  simulate  │
│  glossary│                   │  hyperstit.│                   │  risk tier │
│  social  │                   └──────┬─────┘                   │  phylogeny │
└──────────┘                          │                         └──────┬─────┘
                                      ▼                                │
                               ┌────────────┐                          │
                               │  RECEIPT   │  hash · ledger           │
                               │  FORECAST  │  p only, no Brier        │
                               └──────┬─────┘                          │
                                      ▼                                │
                               ┌────────────┐     ┌──────────┐         │
                               │  SETTLE    │ ──► │  BRIER   │         │
                               │  operator  │     │  series  │         │
                               └────────────┘     └──────────┘         │
                                      │                                │
                    ┌─────────────────┴────────────────┬───────────────┘
                    ▼                                  ▼
             ~/.hyperlex/                       docs/archive/  (Pages)
             receipts · score log               sanitized run history
             vector.db · cache                  not the live DB

Zone	What lives here
Intake	Named routes (`offline` / `live` / `glossary` / `social`) — aliases resolve, offline forces mock
Analysis	Neologisms, 8 lineage families + 2026 YTD leaves, typology, virality, hyperstition stage
Receipt	Integrity-hashed JSON + append-only ledger under `~/.hyperlex/`
Calibration	`pending` → `settle` → `score-series` (only place Brier is real)
Scan / cron	Multi-query `LIVE_EMERGENCE_SCAN` + advisory risk→schedule (never auto-registers)
Research	Phase 5 transmission, multi-agent, risk, phylogeny — always SPECULATIVE
Vector	SQLite or Chroma (local → Cloud promote) for hybrid lineage re-rank
Pages	Static, sanitized history — not your operator store

Why it exists

Most “trend tools” either invent confidence or never close the loop. Hyperlex is built for people who want both edges:

Fast sense-making when a phrase starts climbing
Slow honesty when it’s time to score what you claimed

Phases 0–4 are production skill surface. 5.0–5.3 adds research simulation and operator-loop polish without loosening the settlement rule.


Version	0.4.0
Posture	Hermes skill · Python package (≥3.10) · MIT
Primary store	`~/.hyperlex/`
Public PyPI	Not planned
Abraxas	Wire shapes only — Hyperlex never imports Abraxas

Hard rules

Brier requires settlement — open analysis always has provenance.brier = null.
Phase 5 is SPECULATIVE — sim packets keep brier: null.
No Abraxas hard dependency — hosts may import from Hyperlex.
Local is source of truth — Pages is sanitized history only.
Cron is advisory — risk-schedule proposes jobs; operators register them.

How to use

Install

bash install.sh --dry-run && bash install.sh
export HERMES_SKILL_DIR="${HOME}/.hermes/skills/hyperlex"
export HLX="python3 $HERMES_SKILL_DIR/scripts/hyperlex.py"

$HLX check && $HLX doctor && $HLX smoke
$HLX commands    # simplified map (JSON)

Optional: pip install -e ".[dev]" then python -m hyperlex check.

Daily path (start here) — automatic backend

One command = full results. No manual chaining.

# AUTO: ingest → analyze → receipt → forecasts → score log → Phase 5 risk
$HLX pipeline "rizz" --route offline
# same:
$HLX run "rizz"
$HLX ingest "rizz"                 # full results (use --raw-only for signal only)

# Multi-term bag → one full result per atom automatically
$HLX pipeline "sigma rizz locked in"

# When network is allowed
$HLX pipeline "agentic slop" --route live

Settlement is the only manual step: pending → settle → score-series.

Ingest routes

Prefer routes over raw adapter names:

Route	Meaning	Network
`offline` / `mock` / `default`	Deterministic fixture	no
`live`	Multi-source combine	yes
`glossary`	Betting glossary	yes
`social`	X / Twitter path	yes

$HLX sources
$HLX sources --route live

Aliases: real→glossary · x→x_search · firecrawl→crawl4ai.
HYPERLEX_OFFLINE=1 forces mock for any network source.

Calibration (where Brier is real)

$HLX pending
$HLX settle --forecast-id <id> --decision TRUE   # FALSE | VOID | CONFLICT
$HLX score-series --mean-shift --verify-chain

Scan + advisory cron

$HLX scan \
  --config "$HERMES_SKILL_DIR/examples/cron/scan-queries.json" \
  --route offline --receipt --forecasts --append-log

$HLX risk-schedule --list-tiers
$HLX risk-schedule --tier MODERATE --schedule-out /tmp/hlx-cron

Research (optional)

$HLX simulate --term rizz --mode scenario --domain ai
# Multi-term free text expands to separate atoms (sigma | rizz | locked in)
$HLX terms-split "sigma rizz locked in"
$HLX simulate --term "sigma rizz locked in" --domain ai
$HLX vector-seed --through 2026-08 --include-golden --include-home
$HLX vector-search "rizz" --kind term
$HLX archive-export --include-golden --history

Backfill + populate the vector DB

export HYPERLEX_OFFLINE=1
$HLX lineage-backfill --list --through 2026-08

# Default SQLite
$HLX vector-seed --through 2026-08 --include-golden --include-home
$HLX vector-stats

# Local Chroma (iterate) → promote to Cloud when good
$HLX vector-seed --backend chroma --db ~/.hyperlex/chroma \
  --through 2026-08 --include-home --include-golden
$HLX vector-sync --from-path ~/.hyperlex/chroma --to cloud   # keys in ~/.hermes/.env
$HLX vector-stats --cloud

# One atom per run (do not blend independent terms into one seed)
$HLX run "rizz" --route offline
$HLX run "locked in" --route offline
$HLX run "sharp money" --route offline

Docs: docs/modules/vectordb.md

Local data

~/.hyperlex/receipts/
~/.hyperlex/receipt_ledger.jsonl
~/.hyperlex/score_log.jsonl
~/.hyperlex/cache/
~/.hyperlex/vector.db              # SQLite vectors
~/.hyperlex/chroma/                # local Chroma

Week-one: install → offline run → pending / settle / score-series → only then live routes or higher risk tiers.

Deep dive: docs/operator-loop.md · docs/commands.md

CLI surface

Prefer

Command	Purpose
`commands`	Simplified map
`run "<q>" --route offline`	Daily one-shot
`pending` · `settle` · `score-series`	Calibration
`scan` · `risk-schedule`	Monitoring / cron envelopes
`sources` · `ingest` · `analyze`	Routing & partial pipeline

Also: doctor · relay · signal · diagram · simulate · vector-* · lineage-* · archive-* · ledger-* · smoke

Package subset: python -m hyperlex …

Library

from hyperlex import (
    detect_memetic_patterns,
    pick_source,
    emit_receipt,
    extract_forecasts,
    settle_and_log,
    recompute_series,
    run_phase5_scenario,
    NOT_COMPUTABLE,
    PKG_VERSION,
)

src, _ = pick_source(route="offline")
result = detect_memetic_patterns(
    query="rizz",
    ingest_source=src,
    ingest_route="offline",
)
assert result["provenance"]["brier"] is None

API freeze: docs/api-v1.md · hyperlex.API_V1

Optional Abraxas-shaped exports (no Abraxas import):

from hyperlex.compat.abraxas import to_brier_ledger_entry, list_hlx_runes

Status

Area	Status
Hermes skill + CI	Ready
Route-based ingest + `run` / `pending`	Ready
Lineage + YTD backfill / backprop	Ready
Receipts · settle · score-series	Ready
Risk → scan schedule (advisory)	Ready
Phase 5 research sim	Ready
Local vector DB	Ready
Pages run history	Ready
ANN backend	Deferred
Public PyPI	Not planned

ROADMAP.md · STATUS.md · CHANGELOG.md

Docs & examples

Resource	Purpose
docs site	Workbench + run history
docs/operator-loop.md	Recommended burn-in
docs/commands.md	Command map
docs/modules/ingest.md	Routes & provenance
docs/brier-calibration.md	Forecast → settle → Brier
docs/phase5.md	Research simulation
docs/slang-lineages.md	Lineage methodology
docs/hermes-skill.md	Skill model

Examples: examples/cron/ · examples/receipts/golden/ · examples/calibration/ · examples/slang-families/ · examples/case-studies/

Contributing

See CONTRIBUTING.md. Keep the settlement rule and offline-first defaults intact. Prefer small, test-backed PRs.

PYTHONPATH=src pytest -q
bash install.sh --dry-run

License

MIT — LICENSE.

Built for operators who want memetic signal with a receipt trail, not a mood board.

Name		Name	Last commit message	Last commit date
Latest commit History 145 Commits
.github		.github
assets		assets
data		data
docs		docs
examples		examples
references		references
schemas		schemas
scripts		scripts
src/hyperlex		src/hyperlex
tests		tests
.gitignore		.gitignore
ARCHITECTURE.md		ARCHITECTURE.md
CHANGELOG.md		CHANGELOG.md
CONTRIBUTING.md		CONTRIBUTING.md
DESIGN.md		DESIGN.md
LICENSE		LICENSE
MANIFEST.in		MANIFEST.in
QUICKSTART.md		QUICKSTART.md
README.md		README.md
ROADMAP.md		ROADMAP.md
SKILL.md		SKILL.md
SPEC.md		SPEC.md
STATUS.md		STATUS.md
VERSION		VERSION
hyperlex.manifest.yaml		hyperlex.manifest.yaml
install.sh		install.sh
mkdocs.yml		mkdocs.yml
pyproject.toml		pyproject.toml
skills.sh.json		skills.sh.json

Navigation Menu

Repository files navigation

Hyperlex

Plain English

Atlas

Why it exists

Hard rules

How to use

Install

Daily path (start here) — automatic backend

Ingest routes

Calibration (where Brier is real)

Scan + advisory cron

Research (optional)

Backfill + populate the vector DB

Local data

CLI surface

Library

Status

Docs & examples

Contributing

License

About

Releases

Packages

Contributors

Languages

Folders and files

Latest commit

History

Repository files navigation

Hyperlex

Plain English

Atlas

Why it exists

Hard rules

How to use

Install

Daily path (start here) — automatic backend

Ingest routes

Calibration (where Brier is real)

Scan + advisory cron

Research (optional)

Backfill + populate the vector DB

Local data

CLI surface

Library

Status

Docs & examples

Contributing

License

About

Topics

Resources

Contributing

Security policy

Stars

Watchers

Forks

Releases

Packages

Contributors

Languages