.gitignore

# Logs
logs
.env
STATUS.md
left.md
.audit
performance-audit-report.md
FIXES.md
/.env
python-installer.pkg
Sales20260405-60-494j8t.csv
/Sales20260405-60-494j8t.csv
/natively
natively/
.natively
settings.json
natively/pro_licenses.json
env
/env
*.log
/Users/evin/natively-cluely-ai-assistant/natively
natively
npm-debug.log*
yarn-debug.log*
yarn-error.log*
lerna-debug.log*
.pnpm-debug.log*
release
/release
.release
demo.mp4
tmp
.cursorrules
.windsurfrules
change.md
# nativelyapi is a separate git repo — exclude from parent repo tracking
/nativelyapi

# Local premium dev copy — never commit to public repo
/premium

# Generated migration patches — private, not for public repo
*.patch
# Exception: the patches/ directory holds patch-package commits that MUST
# be checked in (they re-apply automatically in postinstall). This line
# keeps them tracked.
!/patches/*.patch

# Diagnostic reports (https://nodejs.org/api/report.html)
report.[0-9]*.[0-9]*.[0-9]*.[0-9]*.json

# Runtime data
pids
*.pid
*.seed
*.pid.lock

# Directory for instrumented libs generated by jscoverage/JSCover
lib-cov

# Coverage directory used by tools like istanbul
coverage
*.lcov

# nyc test coverage
.nyc_output

# Grunt intermediate storage (https://gruntjs.com/creating-plugins#storing-task-files)
.grunt

# Bower dependency directory (https://bower.io/)
bower_components

# node-waf configuration
.lock-wscript

# Compiled binary addons (https://nodejs.org/api/addons.html)
build/Release

# Dependency directories
# Both spellings on purpose: the trailing slash matches only a DIRECTORY, so in
# a git worktree — where node_modules is usually a symlink to the shared
# install — `git add -A` happily stages it as a 120000 symlink entry pointing at
# an absolute local path, which breaks every other checkout. The slash-less
# pattern also matches the symlink.
node_modules
node_modules/
jspm_packages/
dist-electron/
dist/

# Compiled Apple Speech helper. Built explicitly for local development and
# injected per target architecture during electron-builder afterPack.
/resources/apple-speech/natively-apple-speech

# husky auto-generated machinery — only the user-facing hook files
# (e.g. .husky/pre-commit) are committed.
.husky/_

# Snowpack dependency directory (https://snowpack.dev/)
web_modules/

# TypeScript cache
*.tsbuildinfo

# Optional npm cache directory
.npm

# Optional eslint cache
.eslintcache

# Optional stylelint cache
.stylelintcache

# Microbundle cache
.rpt2_cache/
.rts2_cache_cjs/
.rts2_cache_es/
.rts2_cache_umd/

# Optional REPL history
.node_repl_history

# Output of 'npm pack'
*.tgz

# Yarn Integrity file
.yarn-integrity

# dotenv environment variable files
.env
.env.development.local
.env.test.local
.env.production.local
.env.local

evinresume.pdf
TPA Campus JD - SalesCode.pdf


# parcel-bundler cache (https://parceljs.org/)
.cache
.parcel-cache

# Next.js build output
.next
out

# Nuxt.js build / generate output
.nuxt
dist

# Gatsby files
.cache/
# Comment in the public line in if your project uses Gatsby and not Next.js
# https://nextjs.org/blog/next-9-1#public-directory-support
# public

# vuepress build output
.vuepress/dist

# vuepress v2.x temp and cache directory
.temp
.cache

# vitepress build output
**/.vitepress/dist

# vitepress cache directory
**/.vitepress/cache

# Docusaurus cache and generated files
.docusaurus

# Serverless directories
.serverless/

# FuseBox cache
.fusebox/

# DynamoDB Local files
.dynamodb/

# TernJS port file
.tern-port

# Stores VSCode versions used for testing VSCode extensions
.vscode-test

# yarn v2
.yarn/cache
.yarn/unplugged
.yarn/build-state.yml
.yarn/install-state.gz
.pnp.*

# Swift / SPM
.build/
.swiftpm/
*.xcodeproj
*.xcworkspace

# Generated Proto files
native-audio-service/Protos/
native-audio-service/Sources/CopilotAudioService/Protos/

# OS
.DS_Store

# Electron release builds
release/

# Rust
target/
native-module/target/
native-module/*.node
# Windows locks a loaded .node against overwrite, so build-native.js moves the
# previous artifact aside before rebuilding (see the win32 branch there). The
# leftover is deleted once nothing holds it — usually on the next build.
native-module/*.node.stale-*

.github/RepoGuide.md

# AI / Agent tooling
.sisyphus/
.opencode/
.planning/
opencode.json
.superpowers/
.superbrains/
docs/superpowers/

# Agent documentation patterns
**/AGENTS.md
**/CLAUDE.md
**/.claude/

# Temporary working files
temp/
planning/live-negotiation/temp/

# IDE
/.idea/

# Local test and temporary files
*-test.js
test-*.js
test_*.js
test_*.py
qa-report.json
templog*.txt
dodollm.txt
temp_code_expansion/
cluelyprompts/
# Added by code-review-graph
.code-review-graph/
pitch-deck.html
/pitch-deck.html
Pitch Deck.html
Cluely.html
/Cluely.html
scratch_clean.js
scratch_clean2.js
scratch_extract.js
scratch_svg.html
.agent
.agents
.claude
.codex
diff_settings.txt
PERF_AUDIT.md
refactor_profile.py
refactor.js
skills-lock.json
windowswork.md
scratch/

# Scratch, prompt, and temporary files moved to foxinthehole/
foxinthehole/
# Merge-conflict / backup artifacts (e.g. LLMHelper.ts.orig — never commit)
*.orig

# Report and documentation files (dev-only)
SESSION_MEMORY_ROLLOUT_REPORT.md
AUTOPILOT_PROFILE_INTELLIGENCE_RELEASE_REPORT.md
BEST_IN_CLASS_PROFILE_INTELLIGENCE_FINAL_REPORT.md
BEST_IN_CLASS_PROFILE_INTELLIGENCE_PLAN.md
COMPETITOR_PARITY_AND_SUPERIORITY_REPORT.md
CONTEXT_ARCHITECTURE_DESIGN.md
CONTINUOUS_BENCHMARKING_CI_REPORT.md
ELECTRON_RUNTIME_TEST_REPORT.md
FINAL_LIVE_SESSION_MEMORY_INTEGRATION_REPORT.md
FINAL_MANUAL_PATH_POLISH_REPORT.md
FINAL_RELEASE_HARDENING_REPORT.md
LIVE_MANUAL_SESSION_MEMORY_REPORT.md
LIVE_SESSION_REPLAY_REPORT.md
LIVE_WTA_SESSION_MEMORY_REPORT.md
MANUAL_APP_RETEST_SCRIPT.md
MANUAL_PROFILE_INTELLIGENCE_BUG_REPORT.md
MANUAL_UI_TRUTH_TEST_REPORT.md
OBSERVABILITY_RELEASE_REPORT.md
PROFILE_INTELLIGENCE_BENCHMARKING.md
PROFILE_INTELLIGENCE_FIX_PLAN.md
PROFILE_INTELLIGENCE_FIX_REPORT.md
PROFILE_INTELLIGENCE_IMPROVEMENT_REPORT.md
PROFILE_INTELLIGENCE_LATENCY_TRACE.md
PROFILE_INTELLIGENCE_MANUAL_TESTS.md
PROPOSED_ARCHITECTURE.md
PROVIDER_RESILIENCE_FINAL_REPORT.md
PROVIDER_RESILIENCE_REPORT.md
PROVIDER_TTFT_RACE_REPORT.md
THINKING_BUDGET_BENCHMARK.md
TRANSCRIPT_AWARE_INTENT_ROUTING.md
WHAT_TO_ANSWER_FIX_REPORT.md
reportforai.md
REPORT_TO_CHATGPT.md
REAL_WORLD_E2E_TEST_REPORT.md
REAL_UI_TESTING_APPROACH.md
REAL_UI_EVAL_REPORT.md
REAL_MANUAL_PATH_REGRESSION_REPORT.md
REAL_API_EVAL_REPORT.md
LATENCY_ROOT_CAUSE_REPORT.md
INTELLIGENCE_CONTEXT_ARCHITECTURE.md
INTELLIGENCE_COST_REPORT.md
INTELLIGENCE_MANUAL_TESTS.md
dodopayments.llm
ANSWER_QUALITY_FINAL_REPORT.md
ANSWER_QUALITY_EVAL_REPORT.md
profile.md
# NOT ignored: electron-builder.signed.cjs is the production build config that
# `npm run dist:signed` loads. It was swept into this block of scratch reports and
# then deleted by the "remove tracked files per updated .gitignore" cleanup, which
# broke every signed build with ENOENT. It holds NO secrets — the notarytool
# credential lives in the macOS keychain; only the profile NAME and the (public)
# Team ID appear in the file, and the signing identity is auto-discovered.
REMAINING_ISSUES_FIX_REPORT.md
LONG_SESSION_MEMORY_TEST_REPORT.md
MANUAL_CHAT_PROFILE_FIX_REPORT.md
MULTIMODE_1000_EVAL_REPORT.md
FINAL_FOLLOWUP_STABILITY_RELEASE_REPORT.md
FINAL_RELEASE_REGRESSION_REPORT.md
FOLLOW_UP_RESOLVER_FIX_REPORT.md
WHAT_TO_ANSWER_RELEASE_REPORT.md
FINAL_PROFILE_INTELLIGENCE_RELEASE_REPORT.md
_verify_backend_architect.md
_verify_code_review.md
_verify_test_engineer.md
\# Natively Intelligence OS — Phase-by-Ph.md
ANSWER_DIVERSITY_REPORT.md
BACKGROUND_JOB_ISOLATION_REPORT.md
FINAL_PROFILE_INTELLIGENCE_PRODUCT_QUALITY_REPORT.md
LATENCY_DEGRADATION_REPORT.md
MANUAL_SEQUENTIAL_STRESS_REPORT.md
NATIVELY_EXTERNAL_RESEARCH_NOTES.md
NATIVELY_INTELLIGENCE_OS_E2E_VERIFICATION_REPORT.md
NATIVELY_INTELLIGENCE_OS_FINAL_REPORT.md
NATIVELY_INTELLIGENCE_OS_IMPLEMENTATION_PLAN.md
NATIVELY_INTELLIGENCE_OS_LIVE_VERIFICATION.md
NATIVELY_INTELLIGENCE_OS_OBSERVABILITY.md
NATIVELY_INTELLIGENCE_OS_ROLLOUT.md
NATIVELY_INTELLIGENCE_OS_WIRING_FINAL_REPORT.md
NATIVELY_INTELLIGENCE_OS_WIRING_STATUS.md
natively-intelligence-e2e-results.json
natively-intelligence-e2e-verdict.json
REAL_LONG_SESSION_MANUAL_TEST_REPORT.md
Untitled-1.md

# Eval, benchmark, and test directories
intelligence-eval*/results/
intelligence-eval/
intelligence-eval-answer-quality/
intelligence-eval-real-api/
intelligence-eval-real-ui/
intelligence-eval-results/
intelligence-eval-routing-200/
tempp/
test-results/
profileresume/
playwright-report/
negotiation-routing-eval/
benchmarks/
.playwright-mcp/


# External research clones (not vendored)
_external_research/

# Filled-in credentials intake (never commit real values)
CREDENTIALS-INTAKE.local.md

# Local sprint reports / scratch notes / leaked-prompt reference (not vendored)
NATIVELY_*_REPORT.md
CORE_PIPELINE_AUDIT_REPORT.md
reportforgpt.md
PHASE_STATUS.md
meetily.md
testengineer.prompt
cluely/
docs/

# Large model binary files
*.onnx
*.onnx_data

# Round-7 cleanup (2026-07-01, hardened after senior review): large
# reference PDFs and debug artifacts should NEVER be committed. Keep
# patterns narrow — broad patterns like `Sample*.pdf` and `*thesis*.pdf`
# would silently ignore legitimate PDFs (invoices, resumes, etc.). Use
# the exact filename + the debug-artifacts directory + a single specific
# pattern for the `Sample thesis for testing.pdf` filename family.
debug-artifacts/
Sample thesis for testing.pdf
"Sample thesis for testing.pdf"

# Live-smoke harness scratch dir (created at runtime by scripts/smoke-document-grounded.js)
/.smoke-sandbox/

# User-submitted crash reports / debug logs (device + session diagnostics — never commit)
user log/
*.ips

# --- branch-integration safety hardening (2026-07-23) ---
# Third-party dispute / billing evidence dirs — never push
/dispute-evidence-*/
/dispute-evidence-*
/dispute_evidence/
scripts/*evidence*.py
scripts/combine_*_single_file.py
# The dispute handover docs are ALSO written to the repo root, outside every
# directory rule above — byte-identical strays that a `git add -A` would sweep
# in. This repo is PUBLIC and those files carry a named customer's email, phone
# number and home address, so the root copies need their own rule.
/*_DISPUTE_HANDOVER.md
/*_DISPUTE_*.md
# Autopilot campaign prompts — internal docs, not source
/HANDOFF*.md
/NATIVELY_CONTEXT_SYSTEM_SIMPLIFICATION_PROMPT.md
# Evidence/data-access briefing for investigation agents. Carries no secret
# VALUES, but in a PUBLIC repo it is still a map of the admin surface: which
# /admin/* endpoints exist, that x-admin-secret gates them, the prod API and
# Dodo base URLs, the Supabase table names, and which .env variable unlocks
# each one. That is reconnaissance, so it stays local.
/NATIVELY-EVIDENCE-ACCESS-REFERENCE.md
# Generated Vite build output that mirrors src/**/*.{tsx,ts}
/src/**/*.js

# --- User requested exclusions (2026-07-23) ---
PRODUCTION_HARDENING_FINAL_REPORT.md
loop.md
loop2.md
AUDIT_AI_CONTEXT_OWNERSHIP.md
AUDIT_ARCHITECTURE_MAP.md
AUDIT_AUDIO_REALTIME_PIPELINE.md
AUDIT_BACKEND_SECURITY_STABILITY.md
AUDIT_DEPENDENCIES_PACKAGING.md
AUDIT_ELECTRON_SECURITY.md
AUDIT_ERROR_OBSERVABILITY.md
AUDIT_FRONTEND_STATE_REACT.md
AUDIT_MEMORY_AND_LOOPS.md
AUDIT_PRIVACY_SECRETS_LOGGING.md
AUDIT_VERIFICATION_REPORT.md
campaign-log.md
campaign2-log.md
campaign3-log.md
contextauthresearchdoc.md
test/
test-fixtures/
# …except the two tiny documents SafeDocumentTextExtractor needs. The rest of
# test-fixtures/ is a 46MB corpus regenerated with live MiniMax calls, which no
# CI run can reproduce — so those two suites failed on every run for want of a
# 3.7KB PDF and a 2.2KB DOCX. These are generated deterministically by
# scripts/e2e/lib/docwriters.mjs (jsPDF + JSZip, no network, no model) and carry
# only the marker names the tests assert on.
!test-fixtures/profiles/p01/resume.pdf
!test-fixtures/profiles/p03/resume.docx

traces/
traces2/
traces3/
--help/
.remember/
fixtures/
# ...but NOT test fixtures that suites import. The blanket `fixtures/` above swept up
# electron/**/__tests__/fixtures/*, which the "remove tracked files per updated
# .gitignore" cleanup then deleted — leaving 4 test files importing modules that no
# longer existed. Both lines are required and must stay in this order: git will not
# descend into an excluded directory, so the directory itself has to be re-included
# before its contents can be.
!electron/**/__tests__/fixtures/
!electron/**/__tests__/fixtures/**
# Same sweep deleted tests/fixtures/modes/* (73 files) in 6f8e3329 — the mode
# reference-file corpus that CustomModes / ModePersonaScenarios / Seminar /
# ModeFixtureIntegrity / ModeBleedingMatrix import. That turned ~120 suite
# cases into unconditional ENOENT failures on every fresh checkout. Restored
# from 6f8e3329~1 (2026-08-11) and re-included here so the cleanup cannot
# recur. Same two-line pattern as above: re-include the directory first, then
# its contents.
!tests/fixtures/
!tests/fixtures/**
tools/

# Vision benchmark source documentation + generated reports
!docs/
docs/*
!docs/VISION_MODEL_BENCHMARK.md
# The local embedding model benchmark (2026-09-21/22): code comments across
# electron/rag cite its sections (bundled-model choice, catalog recipes, caps).
!docs/local-embedding-benchmark.md
# Per-release notes are published artifacts (linked from the README), not scratch notes.
!docs/releases/
!docs/releases/**
# Usage ledger — the progress file is the mandated handoff contract across three
# campaigns (16 architectural decisions, 5 open findings, the full forensics
# record), the architecture doc is the system's specification, and the privacy
# diff is a release gate. Untracked, none of it survives a fresh clone or a
# worktree — which is exactly where the next person picks this up.
!docs/usage-ledger-architecture.md
!docs/usage-ledger-progress.md
!docs/usage-ledger-privacy-diff.md
# TypeScript 7 upgrade — audit + report are deliverables, not scratch notes,
# so they are allowlisted out of the blanket `docs/*` ignore above.
# Auto Answer V3 campaign — the progress file is the resume contract for the
# autopilot (a fresh session must be able to continue from it alone) and the
# two specs govern every phase. Same reasoning as the usage-ledger entries.
!docs/autopilot/
!docs/autopilot/auto-answer-v3-progress.md
!docs/specs/
!docs/specs/auto-answer-v2-spec.md.md
!docs/specs/auto-answer-v3-amendments.md
# Reference-retrieval investigation + fix plan (2026-08-28 beta report). These
# are the handover contract for the next agent — a fresh session must be able to
# pick the work up from these four files alone, and a worktree agent cannot see
# an ignored path at all. Same reasoning as the autopilot entries above.
!docs/retrieval-handoff/
!docs/retrieval-handoff/**
# Homebrew cask — the distribution runbook for the tap (issue #577). It carries
# the two steps that cannot be automated from this repo (creating the tap repo,
# adding HOMEBREW_TAP_TOKEN) and the tag-case constraint the cask URL depends
# on. Untracked, a fresh clone ships a cask nobody knows how to publish.
!docs/homebrew-cask.md
benchmark-results/


# Role Insight visual harness — generated artifacts (regenerate with
# `electron scripts/dump-role-insight-fixture.mjs` + `node scripts/role-insight-harness/build.mjs`)
/scripts/.role-insight-fixture.json
/scripts/role-insight-harness/dist/

# Unbuilt feature PRDs — planning documents, never shipped to the public repo
/upcoming/

# Working-directory artifacts that are NOT part of this application.
# These accumulated as untracked noise; none of them belong in the app repo.
#   natively-control/    a separate 1.8 GB / 41k-file project checkout
#   chargeback_evidence/ payment-dispute documents (business records, not code)
#   undefined/           written by a bug that resolved a path to the literal
#                        string "undefined" — if this reappears, fix the caller
#   natively_debug.size* one-off debug dumps
#   pitch/               investor pitch deck + DPIIT notes (business material,
#                        and this repo is public — a deck does not belong in it)
#   researc              a personal chat transcript, saved by a shell redirect
#                        that lost its filename. Kept locally, never committed.
/natively-control/
/chargeback_evidence/
/undefined/
/natively_debug.size*
/pitch/
/researc

benchmarks/reranker-eval/results/

# Reranker/embedding evaluation artefacts and non-shipped weights.
# bge-reranker-large is NOT in scripts/download-models.js's manifest — it is a
# benchmark-only model (~553MB) and must never enter the repo or a build.
resources/models/Xenova/bge-reranker-large/
embedding-benchmark/

# Dev-mode model cache. transformers.js is given `env.cacheDir = <cwd>/models`
# when the app is unpackaged, so running a local reranker or embedder in dev
# downloads weights here — 125MB appeared during one benchmark run. The three
# tracked JSON files under models/Xenova/bge-reranker-base are force-added and
# stay; everything else is a redownloadable cache that must never be committed.
models/*
!models/Xenova/
models/Xenova/*
!models/Xenova/bge-reranker-base/
models/Xenova/bge-reranker-base/*
!models/Xenova/bge-reranker-base/*.json

# Router benchmark training checkpoints (PyTorch .pt, up to 570 MB each). Artifacts, never source.
scripts/intent-benchmark/trained/

# Router benchmark SLM baseline: Qwen3-0.6B Q4_0, 409 MB in ONE file — well over
# GitHub's 100 MB blob limit, so a stray `git add -A` here does not just bloat the
# repo, it produces a history that cannot be pushed at all. Downloaded for the
# `slm-qwen3-06b` provider in scripts/intent-benchmark/run.mjs and measured at p95
# 939ms (66x the leading encoder), i.e. evidence for a rejected approach, never a
# shipped dependency. Nothing under this directory is tracked.
resources/models/gguf/

# Router benchmark model downloads (NLI, GLiClass, potion, bge-small). Weights, never source.
resources/models/MoritzLaurer/
resources/models/Xenova/nli-deberta-v3-base/
resources/models/Xenova/nli-deberta-v3-small/
resources/models/Xenova/bge-small-en-v1.5/
resources/models/knowledgator/
resources/models/minishlab/

# Local scratch: CDP/Playwright session drivers and a campaign's reference-doc pack.
/cdp-*.mjs
/live-*.mjs
/cluely_reference_pack_fresh/

# Local intent-benchmark reports that were never committed (the classifier was removed 2026-09-05).
# Exact names only: the rest of reports/ is tracked.
scripts/intent-benchmark/reports/head-deberta-cs.json
scripts/intent-benchmark/reports/head-deberta-fixed.json
scripts/intent-benchmark/reports/head-deberta-q8.json
scripts/intent-benchmark/reports/hybrid-rules-proto-head-cs.json
scripts/intent-benchmark/reports/hybrid-tight-cs.json
scripts/intent-benchmark/reports/hybrid-wide-cs.json
scripts/intent-benchmark/reports/none-always-answer-cs.json

# Local-embedding bake-off binary output (2026-09-21, docs/local-embedding-benchmark.md).
# The per-model sqlite-vec indexes are ~18MB of regenerable vectors — this repo
# has already had one commit sweep hundreds of MB of model binaries into its
# history, so they are ignored by path. The JSON/CSV results beside them are
# small and deliberately NOT ignored.
results/indexes/
results/live-fixtures/

# Embedders no longer bundled: multilingual-e5-base (bundled 2026-09-21 only,
# never released) and all-MiniLM-L6-v2 (bundled until 2026-09-21; unshipped
# 2026-09-22). A local copy may remain from an older checkout; the R&D baseline
# uses its own pinned copy under the embedding-experiments cache.
resources/models/Xenova/multilingual-e5-base/
resources/models/Xenova/all-MiniLM-L6-v2/

# Bundled multilingual-e5-small (default since 2026-09-22): only the pinned
# manifest is tracked; download-models.js fetches every file by sha256.
resources/models/Xenova/multilingual-e5-small/*
!resources/models/Xenova/multilingual-e5-small/manifest.json

# 9Router provider — the design records endpoint behaviour verified against a
# live instance (auth split by HTTP verb, capabilities carried on /v1/models)
# that is not recoverable from the vendor's docs, plus the reasoning behind the
# `ninerouter` internal id. A later session changing this provider needs it.
!docs/ninerouter-provider-design-2026-09-20.md
# Vision-capability design (2026-10-01). Only this plan is un-ignored; other
# local docs/plans files stay ignored.
!docs/plans/
docs/plans/*
!docs/plans/2026-10-01-vision-capability-design.md
!docs/plans/2026-10-01-vision-capability-phase1-plan.md
!docs/plans/2026-10-01-vision-capability-phase2-plan.md
!docs/plans/2026-10-01-vision-capability-phase3a-plan.md
!docs/plans/2026-10-01-vision-capability-phase3b-plan.md

# Live RAG probe scratch userdata — the probes mkdir these and the app fills
# them with a multi-MB natively.db. Generated output, not a fixture.
tests/e2e-modes/live-userdata/
tests/e2e-modes/hosted-userdata-*/

# Written by `npm run dev:agent` — per-worktree CDP port + session id.
agent-browser.json