Skip to content
transilienceaiPublic

About

Open-source agent for people research, builds psychological portraits from public signals and crafts approaches calibrated to the person and your intent.

Resources

Stars

5 stars

Watchers

0 watching

Forks

Latest commit

 

History

4 Commits

Folders and files

Repository files navigation

Rysy

Built by Transilience MIT License GitHub stars Claude AI

An open-source AI agent that does the homework before you reach out to anyone — researches people, builds psychological portraits, and crafts approaches calibrated to the person and your intent — from the team at Transilience.ai

Quick Start | Personas | Architecture | Customization | Website


Overview

Rysy is an AI agent that researches people using browser automation (LinkedIn, the web), builds psychological portraits from public signals, and helps you craft an approach calibrated to the person and your intent — whether you're pitching an investor, applying to a job, preparing for a podcast interview, or reaching out to a customer.

The core pipeline: research → profile → strategize → witness review.

Same engine, different outputs — the intent is the variable.

Why Rysy?

  • Deep research, not templates — drives Claude in Chrome through your logged-in LinkedIn and the web to read real signals
  • Psychological portraits — synthesizes research into how someone thinks, what they care about, and what would earn their attention
  • Intent-driven output — the same pipeline produces investor cold notes, interview questions, networking briefs, or outreach drafts depending on your goal
  • Built-in quality control — every draft is reviewed by an independent witness agent that doesn't know it's reviewing your work
  • Accumulating craft — learns from every campaign through notes, patterns, and trend tracking
  • Fully customizable — rewrite the constitution, research playbooks, and voice palette for your domain

🎯 Who it's for

Persona Scenario What Rysy produces Why it beats doing it manually
Founder raising a round Researching investors before cold outreach Investor profile (thesis, recent bets, stated vs. revealed preferences), tailored angle, warm-intro path analysis, draft cold note Reads the investor's actual activity — reposts, comments, arguments — and finds the gap between their marketed thesis and their revealed one
Job seeker Understanding a hiring manager before applying or interviewing Hiring manager portrait, team context (recent hires, departures, open problems), interview angle, cover letter draft Turns "why I'm a fit" into "here's the problem you're solving and why my background maps to it"
Sales or BD rep Customer outreach to senior buyers Prospect portrait, objection map, approach strategy (angle, tone, timing, CTA), draft outreach with witness review Deep research → personalized outreach instead of templating at volume
Journalist or podcaster Prepping for a guest interview Guest dossier (career arc, positions, intellectual evolution), prior interview digest, question bank ranked by novelty Surfaces the thread nobody has asked them to pull on yet — the question that makes them lean forward
VC or angel investor Evaluating a founder before or after a pitch Founder profile (builder evidence, leadership signals), narrative consistency check, reference-call prep questions Catches what a LinkedIn skim misses — commit history vs. "technical founder" claims, team churn the deck omits
Conference attendee Prepping for an event with a speaker or attendee list Priority-ranked briefs, conversation hooks per person, connection graph, personalized follow-up templates Fifteen people researched overnight instead of five minutes of panic-Googling between sessions
Recruiter Sourcing and reaching out to senior candidates Candidate portrait (trajectory, motivational signals, move-readiness), role-fit mapping, outreach draft Produces outreach that reads like someone who actually looked — because the agent actually did

🚀 Quick Start

1. Clone and open

git clone https://github.com/transilienceai/Rysy.git
cd Rysy
claude    # Launch Claude Code

2. Install Claude in Chrome

Install Claude in Chrome for browser-driven research. Without it, the researcher falls back to WebFetch and WebSearch for publicly available information.

3. Populate market context

/refresh-trends

4. Run your first campaign

/run-campaign path/to/your-campaign.json

First time setting up? See LOCAL_SETUP.md for the full walkthrough — installing Claude Code, trusting the hooks, granting permissions, and connecting Claude in Chrome.


⚙️ How It Works

You give Rysy a list of people (with LinkedIn URLs) and an intent. Rysy runs four sub-agents:

Agent Role
Researcher Drives Claude in Chrome through your logged-in LinkedIn session and the broader web. Loads playbooks from craft/research-methodology/. Writes structured research notes.
Profiler Reads the research notes in a separate context (so the synthesis isn't colored by the main agent's voice) and writes a psychological portrait.
Rysy Reads the portrait and crafts an approach strategy + draft output calibrated to the person and your stated intent.
Witness Reviews each draft from a stranger's perspective. Returns ship | rewrite | flag with prose. The witness does not know it is reviewing your work — that isolation is the point.

📥 Input

A campaign input has three blocks: campaign (your intent and constraints), sender (who you are), and leads[] (the people to research).

{
  "campaign": {
    "id": "2026-08-series-a-investors",
    "name": "Series A investor outreach",
    "intent": "investor-outreach",
    "context": "We're raising a Series A for our developer tools platform. $2M ARR, 40% MoM growth.",
    "desired_outcome": "Get a first meeting to walk through our metrics and product roadmap",
    "output_format": "cold-note",
    "constraints": {
      "tone": "founder-to-investor, direct, no pitch-deck language",
      "max_length_sentences": 5,
      "avoid": ["buzzwords", "flattery", "asking for money in the first email"]
    }
  },
  "sender": {
    "name": "Your Name",
    "title": "Co-founder & CEO",
    "company": "Your Company",
    "company_one_liner": "developer tools platform that cuts CI/CD pipeline time by 60%",
    "credible_claims": ["ex-infra at Stripe", "YC W24", "open-source project with 2k GitHub stars"]
  },
  "leads": [
    {
      "id": "sarah-chen-first-round",
      "person": {
        "name": "Sarah Chen",
        "linkedin_url": "https://linkedin.com/in/sarah-chen-vc",
        "title": "Partner",
        "location": "San Francisco"
      },
      "company": {
        "name": "First Round Capital",
        "domain": "firstround.com",
        "industry": "venture capital",
        "stage": "established"
      },
      "context": {
        "pre_attached_intel": "Led the Series A for DevToolsCo last year. Writes about developer experience on her blog.",
        "disqualify_signals_check": ["fund currently closed", "competitive portfolio conflict"]
      }
    }
  ]
}

Supported Intents

The intent field shapes the entire pipeline — what the researcher prioritizes, what the profiler emphasizes, and what the output looks like.

Intent What it researches
investor-outreach Thesis, portfolio, recent bets, stated vs. revealed preferences
job-application Hiring manager, team trajectory, open problems, culture signals
customer-outreach Pain points, decision-making style, competitive landscape
interview-prep Prior appearances, intellectual positions, unexplored topics
founder-diligence Builder credibility, team signals, narrative consistency
event-networking Batch-research attendees, connection paths, conversation hooks
candidate-outreach Career trajectory, motivational signals, move readiness
custom Define your own research priorities and output format

📤 Output

After running, you'll find:

File Contents
experience/campaigns/{id}/output.json Structured output with drafts, witness verdicts, and review flags
experience/campaigns/{id}/results-summary.md Drafted vs. skipped vs. flagged counts
experience/campaigns/{id}/run-log.md Timestamped event log
experience/campaigns/{id}/what-i-learned.md Rysy's post-campaign reflection
experience/prospects/{lead-id}/ Full per-prospect working folder (research notes, portrait, drafts)

Drafts marked ready_to_send: true passed both the deterministic quality detector and the witness review. Drafts marked human_review_required: true need your eyes.


🏗️ Architecture

Three Memory Layers

Layer Purpose Mutability
self/ Constitution and voice palette Slow-moving. Human-approved changes only via /apply-diff
craft/ Canon, exemplars, personas, psychology, research playbooks, notes, patterns, trends Accumulates freely through use
experience/ Journal, prospects (gitignored), campaigns Written every campaign

Skills

Skill Purpose
run-campaign Top-level orchestration
take-craft-note Atomic dated observation
promote-pattern Promote note clusters to patterns with evidence
refresh-trends Pull and filter current market narratives
apply-approved-diff The only path to change self/character.md
reindex-memory Rebuild INDEX files across the library

Hooks

Six hooks enforce guardrails automatically — self-protection, quality detection, witness triggers, and session lifecycle management.


🔧 Customization

Rysy is designed to be adapted to your domain. Key extension points:

Extension Point What to customize
self/character.md The constitution — who the agent is, not what it does
self/voice-palette/ Register modes (dry-precise, warm-observational, diagnostic, etc.) for different people
craft/personas/ Persona templates for the types of people you interact with
craft/research-methodology/ Research playbooks — what to look for, where to look, how to weigh signals
craft/exemplars/ 5-10 examples of great output for your use case
craft/canon/ The intellectual lineage your agent draws from

🔒 Privacy

experience/prospects/ contains personal data about real people and is gitignored. Treat it accordingly.

Research is conducted through your own logged-in browser session — Rysy sees what you can see. No scraping APIs, no data brokers, no dark patterns. The line between attention and intrusion is part of the constitution.


🗺️ Roadmap

  • Phase 1 (now) — research → portrait → strategy → draft → witness review; output for human review
  • Phase 2 — memory across campaigns; promoted patterns inform future work; cross-campaign retrieval
  • Phase 3 — fuller autonomy: target identification → research → outreach → reply handling → follow-up with guardrails

Project Stats

Category Count
Sub-agents 4 (researcher, profiler, witness, introspecter)
Skills 6
Hooks 6
Supported Intents 8
Research Playbooks 7
Voice Registers 4

Community & Support


License

MIT License — Copyright (c) 2026 Transilience AI. See LICENSE for details.


Contributors


Built by Transilience AI

We build AI-driven cloud security and compliance automation. Rysy is how we think about people research — if you're curious about what else we're building, take a look.

Star this repository

Website | Issues

claude-code ai-agent people-research outreach linkedin browser-automation psychological-profiling investor-outreach job-search sales recruiting multi-agent

About

Open-source agent for people research, builds psychological portraits from public signals and crafts approaches calibrated to the person and your intent.

Resources

Stars

5 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages