TLS-Chameleon

TLS-Chameleon is a modern Python HTTP networking stack with pluggable browser-fingerprint backends, structured fingerprint research tooling, diagnostics, and reproducible experiments — behind a simple, requests-like API.

What it IS: an HTTP client · a fingerprint-aware networking toolkit · a diagnostic system · a protocol-research framework.

What it is NOT: a Cloudflare/WAF bypass guarantee · a CAPTCHA solver · an anonymity or stealth product. Detection outcomes depend on many factors outside any client library's control.

🆕 What's New in v3.0.0

• Pluggable transport architecture — curl, native (primp/rustls) and httpx backends behind one interface, auto-selected (curl → native → httpx) with graceful degradation. curl_cffi is optional, never architectural.
• Structured fingerprint system — typed models, registry over 48 profiles, validator (rejects impossible/inconsistent configs), explainable similarity scoring, field-level diffing, live capture via TLS echo endpoints.
• Diagnostics — response.trace, inspect_url(), doctor() with check-by-check verdicts; every output automatically redacted.
• Adaptive engine — bounded/expiring/thread-safe domain memory with explainable selection (client.profile_for(domain)); header-consistency engine; deterministic randomization via random_seed.
• CLI — chameleon get / inspect / doctor / capture / diff / fingerprint / benchmark / version, all major commands with stable --json.
• Reproducible benchmarks — real local-server harness, stored methodology, no invented numbers.
• Honest capability reporting — client.capabilities.tls_fingerprint_spoofing, .http3, ... always reflect what the active backend actually does.

🚀 Features

• Three interchangeable backends
  • curl — curl-impersonate (curl_cffi): full JA3/JA4/H2 fingerprint control
  • native — primp/rustls stack: browser impersonation without libcurl
  • httpx — honest fallback: standard OpenSSL TLS (no JA3 spoofing)
• 48+ versioned profiles: Chrome, Firefox, Safari, Edge across Windows 10/11, macOS, Linux, iOS, Android — plus a deterministic generative engine (gen://family/os/major/tier/seed)
• Fingerprint research toolkit: registry · validation · similarity · diff · live capture
• Diagnostics & doctor: protocol/backend/timing traces, observed-vs-profile comparison, actionable recommendations
• Adaptive behavior: per-domain profile learning (bounded, expiring, thread-safe, disableable), header casing/order morphing, WAF detection with retry/backoff/rotation
• Deterministic randomization: same seed ⇒ identical variants, cipher order and jitter — reproducible experiments
• Resilience: proxy/profile pools, rate limiting, ghost mode, on_retry hooks
• Magnet module 🧲: emails, tables, forms, JSON-LD, deep extraction of JWTs/API keys; optional AI providers ([ai] extra)

📦 Install

pip install tls-chameleon            # core: works out of the box via httpx
pip install tls-chameleon[curl]      # + curl-impersonate backend (JA3 spoofing)
pip install tls-chameleon[native]    # + primp/rustls backend (JA3 spoofing, no curl)
pip install tls-chameleon[all]       # everything

Backend honesty: without [curl] or [native] you get the httpx fallback — standard OpenSSL TLS, no JA3 spoofing. The active backend and its true capabilities are always inspectable:

from tls_chameleon import TLSSession

client = TLSSession()
print(client.engine)                                  # "curl" | "native" | "httpx"
print(client.capabilities.http3)                      # only if truly available
print(client.capabilities.tls_fingerprint_spoofing)   # False on httpx!

Pluggable architecture

Public API  (TLSSession / AsyncSession — unchanged names since v2)
     │
tls_chameleon.transport.factory      # auto / curl / native / httpx (+ custom)
     │
Transport interface                  # duck-typed sessions, capability reports
 ├─ CurlTransport    ← only module importing curl_cffi
 ├─ PrimpTransport   ← only module importing primp
 └─ HttpxTransport   ← only module importing httpx

Backends are strictly isolated (enforced by tests). Custom backends plug in via tls_chameleon.transport.register_transport.

⚡ Quick Start

from tls_chameleon import TLSSession

with TLSSession(profile="chrome_130_win11") as client:
    r = client.get("https://example.com")
    print(r.status_code, r.text[:80])
    print(client.engine)                 # which backend served this?
    print(client.capabilities.to_dict()) # what can it really do?

Async:

import asyncio
from tls_chameleon import AsyncSession

async def main():
    async with AsyncSession(profile="chrome_130_win11") as session:
        r = await session.get("https://example.com", trace=True)
        print(r.trace.protocol, r.trace.timing_ms)

asyncio.run(main())

CLI:

chameleon inspect https://example.com --json
chameleon doctor https://example.com --echo-endpoint https://tls.peet.ws/api/clean
chameleon capture https://tls.peet.ws/api/all --raw --json
chameleon fingerprint list --browser chrome
chameleon diff capture_a.json capture_b.json

🔬 Fingerprint System

from tls_chameleon import (
    FingerprintRegistry, validate_fingerprint,
    FingerprintSimilarity, diff_fingerprints, capture,
)

reg = FingerprintRegistry()
fp = reg.get("chrome_120_win11")          # lazy lookup over all built-ins

issues = validate_fingerprint(fp)          # structural + provenance checks

result = FingerprintSimilarity().compare(fp, reg.get("firefox_120_win11"))
print(result.total, result.layers)         # explainable, weighted scoring

report = diff_fingerprints(fp, reg.get("firefox_120_win11"))
print(report.to_text())                    # SAME/DIFFERENT per field + score

# Live capture: what does the network ACTUALLY see?
res = capture(session=client.session)      # via TLS echo endpoint
print(res.fingerprint.tls.ja3_hash)        # source="captured", timestamped

Provenance is explicit — every fingerprint is labeled captured, documented or synthetic; synthetic data can never be marked verified (enforced by the validator).

🩺 Diagnostics

from tls_chameleon import inspect_url, doctor

print(inspect_url("https://example.com", client).to_text())

report = doctor("https://example.com",
                echo_endpoint="https://tls.peet.ws/api/clean")
print(report.to_text())
# [ ✓] Connection: h2 response 200 in 76ms
# [ ✓] Backend: backend 'curl' performs real TLS impersonation
# [ ⚠] Fingerprint (JA4): observed JA4 differs from profile ...
# Verdict: WARN

Traces attach to responses on demand — headers always redacted:

r = client.get(url, trace=True)
r.trace.backend / .protocol / .timing_ms / .request_headers

Unobservable fields stay None with an explanatory note — never guessed.

🧠 Adaptive Engine

client = TLSSession(adaptive=True, adaptive_ttl=3600, random_seed=12345)

client.profile_for("example.com")
# {'profile': 'chrome_130_win11', 'reason': 'learned after 3 successful
#  request(s); 12s ago', 'confidence': 0.6, 'last_used': ...}

Domain memory is LRU-bounded, TTL-expiring, thread-safe, stores only domain → profile (never credentials), and explains itself. Same seed + config ⇒ byte-identical fingerprint choices for reproducible runs.

📚 Profiles

chameleon fingerprint list                # or: list_available_profiles()
chameleon fingerprint show chrome_130_win11 --json
chameleon fingerprint validate my_profile.json

Generative fingerprints for research/fuzzing: TLSSession(profile="gen://chrome/win11/124/balanced/7") — deterministic per seed, always labeled synthetic.

🛠 API Reference (selection)

Handy members: client.capabilities, client.profile_for(domain), session.get_fingerprint_info(), response.trace, save_cookies/load_cookies/export_session/import_session, plus the Magnet extractors (response.magnet.*) and submit_form().

🖥 CLI

All major commands accept --json with stable, documented schemas. (*3 = feature pending its phase.)

📊 Benchmarks

Real local-server measurements only — methodology and limitations in docs/BENCHMARK_METHODOLOGY.md, a labeled sample run in docs/BENCHMARK_SNAPSHOT.md. Absolute numbers are machine-specific; compare within a single report.

📖 Documentation

🤝 Contributing

Issues and Pull Requests welcome!

🌟 Credits

Built on curl_cffi, primp, and httpx.

☕ Support / Donate

If you found this library useful, buy me a coffee!

📜 License

MIT

🚨 Is this library failing on a specific site?

Please open an issue with the URL! I need test cases to improve the fingerprinting logic.