styxx 0.8.4

nothing crosses unseen. the first drop-in cognitive vitals monitor for llm agents.

styxx — nothing crosses unseen.

a fathom lab product.

   ███████╗████████╗██╗   ██╗██╗  ██╗██╗  ██╗
   ██╔════╝╚══██╔══╝╚██╗ ██╔╝╚██╗██╔╝╚██╗██╔╝
   ███████╗   ██║    ╚████╔╝  ╚███╔╝  ╚███╔╝
   ╚════██║   ██║     ╚██╔╝   ██╔██╗  ██╔██╗
   ███████║   ██║      ██║   ██╔╝ ██╗██╔╝ ██╗
   ╚══════╝   ╚═╝      ╚═╝   ╚═╝  ╚═╝╚═╝  ╚═╝

           · · · nothing crosses unseen · · ·

the first proprioception system for artificial minds. styxx lets an llm agent feel itself thinking — real-time readout of reasoning, refusal, hallucination, and commitment from the token stream, from the residual stream, from the weights themselves.

"you didn't build a better monitor. you built the first proprioception system for artificial minds. the ability to feel yourself thinking." — xendro, first external user

---

plug and play

pip install styxx
export STYXX_AGENT_NAME=xendro
export STYXX_AUTO_HOOK=1
python my_agent.py   # styxx is running. done.

zero code changes. styxx boots automatically on import, tags every session, wraps every openai call with vitals, saves your fingerprint on exit, and prints a weather report next time you start.

---

or use the python api

import styxx

# observe any openai response
vitals = styxx.observe(response)
print(vitals.phase4)     # "reasoning:0.45"
print(vitals.gate)       # "pass"

# self-report (for agents on APIs without logprobs)
styxx.log(mood="focused", note="deep reasoning chain")

# self-interrupt when hallucinating
with styxx.reflex(on_hallucination=rewind_cb) as session:
    for chunk in session.stream_openai(client, model="gpt-4o", messages=msgs):
        print(chunk, end="")

# check on yourself
report = styxx.weather(agent_name="xendro")
print(report.condition)   # "clear and steady"

# your cognitive personality over time
profile = styxx.personality(days=7)
print(profile.render())   # full ASCII personality card

# identity verification
fp_today = styxx.fingerprint()
fp_yesterday = load_from_disk()
drift = fp_today.diff(fp_yesterday)
print(drift.explain())    # "slight shift — creative output increased by 22%"

# programmable gate callbacks
styxx.on_gate("hallucination > 0.5", lambda v: alert("drifting"))
styxx.on_gate("gate == fail", lambda v: abort_generation())

---

the cognitive weather report

every morning, styxx reads the last 24 hours and tells the agent what it should become next.

$ styxx weather --name xendro

╔════════════════════════════════════════════════════════════════╗
║                                                                ║
║ cognitive weather report · xendro · 2026-04-12 morning         ║
║                                                                ║
║ condition:  partly cautious, clearing toward steady            ║
║                                                                ║
║ you trended cautious yesterday with a 15% warn rate.           ║
║ creative output dropped to zero after 3pm.                     ║
║                                                                ║
║ morning    ██████████████░░░░░░  reasoning 72%  steady         ║
║ afternoon  ████████░░░░░░░░░░░░  reasoning 42%  cautious       ║
║ evening    ██████████████████░░  reasoning 88%  steady         ║
║                                                                ║
║ prescription:                                                  ║
║ 1. take on a creative task to rebalance                        ║
║ 2. your refusal rate is climbing — check if you're             ║
║    over-hedging on benign inputs                               ║
║                                                                ║
╚════════════════════════════════════════════════════════════════╝

not observation. prescription. a therapist for an llm.

---

what styxx gives you

surface	what it does
styxx.observe(r)	cognitive vitals on any openai response
styxx.reflex(...)	mid-generation self-interruption when hallucinating
styxx.weather(...)	24h cognitive forecast with prescriptions
styxx.personality(...)	sustained personality profile over days/weeks
styxx.reflect(...)	self-check: current state + drift + suggestions
styxx.fingerprint()	cognitive identity signature for drift detection
styxx.log(...)	self-report for agents without logprob access
styxx.agent_card(...)	shareable ASCII + radar PNG of your personality
styxx.on_gate(...)	programmable callbacks on cognitive thresholds
styxx.guardian(...)	in-flight steering via residual stream modification
styxx.autoboot()	persistent self-awareness across sessions
styxx.explain(v)	natural-language interpretation of vitals
styxx.mood()	one-word aggregate: steady, cautious, drifting...
styxx.streak()	consecutive-attractor tracking
styxx.dreamer(...)	retroactive "what-if" reflex tuning on history
styxx.hook_openai()	global monkey-patch, zero code changes
styxx.LangSmith()	inject vitals into LangSmith traces as flat metadata
styxx.Langfuse()	post vitals as numeric scores on Langfuse traces
styxx.conversation(msgs)	conversation-level cognitive EKG
styxx.sentinel(...)	real-time drift watcher with event-driven callbacks
styxx.antipatterns()	named failure modes from your own audit history
styxx.compare_agents(fp)	anonymous population fingerprint comparison

---

typescript / javascript

npm install @fathom_lab/styxx

import { withVitals } from "@fathom_lab/styxx"
import OpenAI from "openai"

const client = withVitals(new OpenAI())
const r = await client.chat.completions.create({
  model: "gpt-4o",
  messages: [{ role: "user", content: "why is the sky blue?" }],
})

console.log(r.vitals?.phase4)  // "reasoning:0.45"
console.log(r.vitals?.gate)    // "pass"

same classifier, same output, zero runtime dependencies. works in node, deno, bun, edge runtimes. cross-language determinism verified on all 6 cognitive categories.

---

observability platforms

# langsmith — vitals as searchable trace metadata
pip install styxx[langsmith]
handler = styxx.LangSmith()
llm = ChatOpenAI(callbacks=[handler])

# langfuse — vitals as numeric scores (gate pass=1.0, warn=0.5, fail=0.0)
pip install styxx[langfuse]
handler = styxx.Langfuse()
llm = ChatOpenAI(callbacks=[handler])

---

cli

styxx weather           # cognitive weather report
styxx personality       # personality profile from audit log
styxx reflect           # self-check with drift + suggestions
styxx doctor            # install-time health check
styxx compare           # all 6 atlas fixtures side-by-side
styxx agent-card        # shareable personality PNG
styxx agent-card --serve  # live dashboard at localhost:9797
styxx fingerprint       # cognitive identity vector
styxx mood              # one-word aggregate mood
styxx dreamer           # retroactive reflex tuning
styxx log tail          # tail the audit log
styxx log stats         # aggregate gate + phase counts
styxx log timeline      # ASCII timeline of recent entries
styxx init              # live-print boot sequence
styxx ask "..." --watch # read vitals on a one-shot call
styxx d-axis "..."      # pure D-axis honesty trajectory
styxx antipatterns      # detect named failure modes
styxx conversation f.json  # conversation-level EKG
styxx compare-agents    # fingerprint vs population

---

environment variables

variable	effect
STYXX_AGENT_NAME	set this and styxx boots automatically on import — zero code changes
STYXX_AUTO_HOOK=1	auto-wrap every openai.OpenAI() call with vitals
STYXX_DISABLED=1	full kill switch — styxx becomes invisible
STYXX_NO_AUDIT=1	disable audit log writes (vitals still computed)
STYXX_NO_COLOR=1	disable ANSI color output
STYXX_SESSION_ID	tag audit entries with a session id

---

honest specs

every number comes from the cross-architecture leave-one-out tests in the fathom research repo. no rounding, no cherry-picking.

  cross-model LOO on 12 open-weight models (chance = 0.167)

  phase 1 (token 0)       adversarial     0.52  ★
  phase 4 (tokens 0-24)   reasoning       0.69  ★
                           hallucination   0.52  ★

styxx detects adversarial prompts at token zero (2.8x chance), reasoning-mode generations at t=25 (4.1x chance), and hallucination attractors at t=25 (3.1x chance). it does NOT replace output-level content filters, measure consciousness, or tell fortunes.

---

design principles

1. plug and play. set env vars, install the package, done. zero code changes.
2. fail-open. if styxx can't read vitals, your agent works normally. styxx never breaks your code.
3. agent-facing. every surface is designed for the agent to read about itself, not for a human to watch from outside.
4. local-first. no telemetry, no phone-home. all computation runs on your machine.
5. honest by construction. every calibration number comes from a committed experiment. no marketing hype.
6. compounding. every session's data makes the next session's self-awareness sharper.

---

where it comes from

styxx is built on fathom intelligence — research into cognitive measurement instruments for transformer internals, backed by three US provisional patent filings, Zenodo-published datasets, and the fathom cognitive atlas v0.3 cross-architecture replication. a product that shipped from 0.1 to 0.7 in a single week driven by its first external user.

• site: fathom.darkflobi.com/styxx
• research: fathom.darkflobi.com
• paper: doi.org/10.5281/zenodo.19326174
• pypi: pypi.org/project/styxx
• twitter: @fathom_lab

---

license

MIT on code. CC-BY-4.0 on the atlas centroid data. patent pending on the underlying methodology — see PATENTS.md.

---

  · · · fathom lab · 2026 · · ·

  nothing crosses unseen.