Kluster☣︎Phuchr™
field terminal · sourced strategy operator
turns messy questions and documents into sourced strategies, decisions, scripts, workflows, and downloadable work. One source-backed strategist — not a chatbot, not separate bots. It converts messy questions, documents, and decisions into structured, cited, downloadable work.
Yo Kluster, give me the MICE framework in 2 sentences flat.
People spy for **Money, Ideology, Coercion, or Ego** — that's the whole damn framework 1. Every asset is a cocktail of those four, and spotting which one they're chasing is how you play the game 2.
If you're trying to flip someone, run the recruitment ladder: spot → assess → develop → recruit → handle 1. Most operators skip assessment and get burned. ....*poot*....
0 docs · 0 doctrine domains
Curated, attribution-tagged doctrine. Every retrieval shows you the source, the category, and the relevance score. Hover any node for the breakdown.
MICE · KUBARK · HUMINT cycle · intelligence-collection doctrine.
Behavioral Change Stairway · PEACE interviewing · threat assessment.
Sun Tzu · Clausewitz · OODA · COIN doctrine · maneuver warfare.
Hormozi · SPIN · Challenger · Voss · Camp · Sandler · Tracy.
Tempe / Phoenix intel — neighborhood, transit, vendor context.
Chain-of-custody · evidence handling · prompt-injection defense.
Token architecture · component contracts · stateless UI doctrine.
Free trial — 30 turns + 5 web searches, no card needed
Usage-based, not time-based. Launch the terminal to start your trial; when the turns run out, pick a tier below.
Pick your tier.
free forever
- • 30 turns + 5 web searches
- • 100+ doc knowledge tree
- • Full-text search
- • Source inspector
unlimited everything
- • Unlimited turns
- • All 12 workflows + exports
- • Document ingestion
- • Conversation tags + archive
- • Settings + accent themes
Real prices. No invented tiers. Stripe-secured. Cancel anytime.
One operator. Messy in. Structured out.
Source-backed retrieval
Every answer draws from a curated knowledge tree — CIA tradecraft, FBI behavioral science, war studies, sales frameworks, Phoenix local intel. You see the exact docs pulled and their relevance scores. No hallucinated citations.
Versioned workflows
Strategy briefs, offer construction, negotiation prep, decision analysis — each a schema-bound pipeline: validate input → retrieve doctrine → synthesize → validate output → render → export.
Document ingestion
Upload TXT/MD/JSON/JSONL/CSV/HTML/PDF/DOCX. Admin datasets: hash + provenance → extract → classify → dedupe → PII neutralization (names→generic placeholders, data preserved) → normalize + citation double-check → validate → score → quarantine if uncertain → index only if approved. User uploads: stored as-is.
Downloadable deliverables
Every workflow exports Markdown, JSON, CSV, or TXT — hashed, ownership-bound, retention-managed. Download links work straight from chat and expire on schedule.
Provider resilience
GLM, DeepSeek, and OpenAI under one abstraction. Native streaming, retries, backoff, timeouts, aborts, normalized errors, automatic provider fallback. Never a hard crash.
Turn-level observability
Every turn logs provider, model, prompt hash, retrieval method, query terms, source IDs + scores, confidence, fallback reason, token usage, estimated cost, latency, and feedback. Full admin console.
Type the messy version. Get the sourced version.
Yo Kluster, give me the MICE framework in 2 sentences flat.
People spy for Money, Ideology, Coercion, or Ego — that's the whole damn framework. Every asset is a cocktail of those, and spotting which one they're chasing is how you play the game. ....*poot*....
Structured deliverables, not just chat.
Each workflow is a versioned pipeline with JSON Schema contracts for input, output, and validation. You get clear normal language and a downloadable file — the schemas stay internal.
Strategy Brief
Messy problem → prioritized options, risks, next actions.
Offer Construction
Hormozi value equation → Grand Slam offer + risk reversals.
Negotiation Prep
Interests, BATNA, ZOPA, calibrated questions, black swans.
Decision Analysis
Weighted criteria, option scoring, kill criteria, reversibility.
No hallucinated sources. Ever.
Lexical scoring with fallback doctrine
Every query is tokenized and scored against the full 100+ document knowledge base with field-weighted relevance (title, tags, summary, body × authority weight). If nothing matches, it falls back to high-weight core doctrine — never empty.
Live retrieval inspector
The terminal's inspector panel shows query terms, injected memory docs, relevance scores, latency, and fallback status in real time. You see exactly what fed the answer.
Provenance preserved
Every ingested document keeps its original filename, MIME, byte size, SHA-256 hash, uploader, and source URL. Processing history is recorded for safe reprocessing.
Built for anyone with a hard question.
Founders & operators
Turn a brain-dump into a funded strategy brief and a Grand Slam offer in one pass.
Sales & closers
Walk into the room with a full negotiation prep sheet — BATNA, anchors, labels, black swans.
Researchers & analysts
Interrogate documents with sourced retrieval. Get dossiers, not hallucinations.
Anyone with a hard question
Type the messy version. Get the structured, sourced, downloadable version.
Your data is yours. User uploads stay as-is.
Owner isolation
Conversations, workflow runs, and generated files are bound server-side to a stable per-browser identity. Private user docs never mix into the shared knowledge tree.
Export & delete
Generated files expire on a retention schedule. You can export and delete your data. RAG retrieval is not model training — ever.
Straight answers.
01Is this just another chatbot?
No. Kluster☣︎Phuchr™ is a source-backed strategy operator. Every answer is grounded in a 100+ document knowledge tree spanning doctrine domains (CIA tradecraft, FBI behavioral science, war studies, proven sales frameworks, Phoenix/Tempe local intel, forensics, design systems, philosophy, trial advocacy, behavioral science, legal, system design, temporal) plus any documents you ingest. It produces structured deliverables you can download — not just chat.
02Where does the knowledge come from?
A curated, attribution-tagged knowledge base spanning the Intelligence Cycle, MICE/HUMINT, KUBARK, PEACE interviewing, the Behavioral Change Stairway, Sun Tzu, Clausewitz, the OODA loop, COIN doctrine, Hormozi, Brian Tracy, SPIN, Challenger, Voss, Camp, Sandler, plus Phoenix/Tempe local intel. Every retrieval shows you the sources and their relevance scores.
03What happens to files I upload?
Default: stored as-is — your documents are never modified. PII redaction is OFF by default; toggle it on in Settings to mask structured PII (emails, phones, SSNs, cards, IPs, addresses) using [EMAIL]/[PHONE]/… markers. Ordinary names, framework names, research/source material, and knowledge content are never auto-redacted. Admin dataset inputs (shared retrieval knowledge — never model training data) run neutralizePii() which replaces real names with generic placeholders so no real PII enters the shared knowledge tree, while the data/logic stays fully extractable.
04Do you train models on my data?
No. RAG retrieval is not training. Your documents are stored for your use, retrieved for your conversations, and never used to train or fine-tune any model.
05What can it actually produce?
Strategy briefs, offer stacks, negotiation prep sheets, decision analyses, research dossiers, action plans — each as a readable answer plus downloadable Markdown/JSON/CSV/TXT files with working links, hashes, and retention.
06Is my data private?
User uploads are stored as-is (never modified) unless you opt in to PII redaction. Conversations and ingested docs are tenant-isolated by a stable per-browser identity. Generated files expire on a retention schedule. Raw turn telemetry is auto-deleted after 30–90 days (configurable) — daily aggregates survive. The app does NOT collect IP-derived geography. See the Policies page for the full, source-faithful details.
07What if the AI provider goes down?
A provider abstraction spans GLM, DeepSeek (V3 chat or R1 reasoner), and OpenAI-compatible endpoints with retry, backoff, and automatic provider fallback. The PROVIDERS badge in the top bar is a click-controlled selector — switch immediately, selection persists (cookie for free users, Supabase-synced for paid). If every provider fails, you get an in-character fallback reply — never a hard crash.
08How does the free trial work?
Usage-based, not time-based. Free users get a configurable turn allowance (default 30, server-side, env-configurable) plus 5 free onboarding web searches to experience real web research. There is NO countdown timer — the trial ends when you use up your turns. Clearing cookies does NOT reset the trial (the ledger is keyed by your stable per-browser identity in Supabase). After exhaustion, free use is blocked at the upgrade gate.
09Are the web citations real?
Yes. The /api/web-research pipeline runs REAL web retrieval via z-ai-web-dev-sdk (web_search + page_reader). It fetches actual source pages, extracts evidence, matches each factual claim to a verbatim quoted excerpt from a retrieved source, and renders citations from the actual URLs. No successful web retrieval → the system does NOT claim it searched the internet. No source evidence → no web-derived factual claim. Citation links resolve to the actual source used.
10Is there a chat character limit?
No. The chat textarea accepts massive pasted context with no artificial 8K/8.5K cap. The previous limit was removed per spec — no silent clipping, substringing, summarizing, or partial submission. The full message is stored byte-faithfully and sent to the model in full. The box grows up to 60vh, then scrolls internally.
11Do my settings follow me across devices?
For paid (Operator) users — yes. Settings sync to Supabase (authoritative) with localStorage as the cache/fallback. Log in on a different device and your preferences restore automatically. Free users persist settings to browser localStorage only.