LaunchSim

Rehearse your product launch in a digital sandbox before spending a dollar.

LaunchSim creates a swarm of synthetic personas powered by an LLM and simulates true swarm dynamics: agents influence each other through an accumulated social signal, so when an influencer buys or shares, it raises the odds for every agent that comes after — just like a real launch. Returns conversion metrics, social posts, market objections, and strategic insights — in minutes.

How to use it (3 steps)

Step 1 — Install

git clone https://github.com/dochoar/LaunchSimulation.git
cd LaunchSimulation

npm run setup            # installs frontend dependencies
npm run install:backend  # installs Python backend dependencies
cp backend/.env.example backend/.env

Then open backend/.env and configure your LLM (see LLM options below).

Step 2 — Describe your product

Create an input.txt file at the root of the project (see input_example.txt for reference). Edit the values to match your product:

name=Your Product Name
description=What problem does it solve, and who has that problem?
        How does your product solve it? (key features, not buzzwords)
        Why is the price fair? What makes you different from alternatives?
        Be specific — the more detail, the more realistic the simulation.
price_usd=29.0
channel=social_media
target_market=
num_agents=50

Channel options: social_media · email · paid_ads · seo · word_of_mouth · app_store · other

💡 Tip: Copy input_example.txt to input.txt and modify the values:
cp input_example.txt input.txt

Step 3 — Run

Open two terminals:

# Terminal 1 — start the backend
npm run dev:backend

# Terminal 2 — run your simulation
python simulate.py

That's it. The script reads your product details from input.txt, submits your product, polls until the simulation completes, and prints a full report in the terminal.

To also open the visual dashboard, start the full stack instead:

npm run dev   # starts both backend (8000) and frontend (3000)

Then go to http://localhost:3000 and use the web UI (which also reads from input.txt or lets you fill in a form).

What you get

════════════════════════════════════════════════════════════════
  LaunchSim  ·  Your Product Name
════════════════════════════════════════════════════════════════
  Price      : $29.00
  Channel    : social_media
  Agents     : 50 synthetic personas

  ⠹  RUNNING      personas= 32  interactions=  0  posts=  0

════════════════════════════════════════════════════════════════
  RESULTS
════════════════════════════════════════════════════════════════
  Viewed        :   31  (62.0%)
  Clicked       :   18  (36.0%)
  Purchased     :    4  (8.0%)
  Conversion    : 8.0%
  Avg sentiment : +0.38   (-1 hostile → +1 enthusiastic)

════════════════════════════════════════════════════════════════
  PERSONAS GENERATED (first 5)
════════════════════════════════════════════════════════════════
  agent_001  Marcus Chen            early_adopter    WTP=$49     [BUYS    ]
             Senior engineer at Series B fintech. Drowning in PR review queue...

  agent_002  Priya Nair             pragmatist       WTP=$12     [BOUNCES ]
             Junior dev waiting 3 days for PR feedback. Price is a stretch...

════════════════════════════════════════════════════════════════
  TOP OBJECTIONS
════════════════════════════════════════════════════════════════
  #1  [14x]  Price point too high for individual devs — feels like a team tool
  #2  [9x]   Privacy concerns: proprietary code sent to external AI API

════════════════════════════════════════════════════════════════
  STRATEGIC INSIGHTS
════════════════════════════════════════════════════════════════
  1. Add a $15/mo solo tier — 14 non-buyers cited price as the only blocker
  2. Publish SOC 2 docs — 9 agents hard-blocked by security/privacy concerns
  3. Mentorship Mode is underused in marketing — junior devs love it but don't see it

Writing a good description

The description is the most important field. It feeds the Researcher (who searches competitors and pricing) and the Ethnographer (who generates personas calibrated to your market).

Include:

The specific problem — who has it, how often, how painful
How your product solves it (concrete features, not "AI-powered innovation")
The price and what they get for it
What makes you different from tools they already use

Good:

CodeReview AI posts an automated code review within 60 seconds of opening a PR on GitHub. It catches logic errors, security vulnerabilities (OWASP Top 10), N+1 queries, and explains every issue in plain English with a concrete fix. Includes Mentorship Mode for junior devs. Learns your team's coding standards in 2 weeks. $29/mo per seat, 14-day free trial. For engineering teams of 2–10 drowning in PR review backlog.

Too vague:

An AI tool that helps developers write better code with advanced features and improve their workflow.

LLM options

Edit backend/.env to switch models. Any OpenAI-compatible API works.

Local — Ollama (free, private, no internet required)

LLM_API_KEY=ollama
LLM_BASE_URL=http://localhost:11434/v1
LLM_MODEL_NAME=qwen2.5:7b

LLM_BOOST_API_KEY=ollama
LLM_BOOST_BASE_URL=http://localhost:11434/v1
LLM_BOOST_MODEL_NAME=qwen2.5:7b

Pull the model first:

ollama pull qwen2.5:7b

50 agents on a mid-range CPU: ~5–12 minutes.

Cloud — OpenAI

LLM_API_KEY=sk-your-key-here
LLM_BASE_URL=https://api.openai.com/v1
LLM_MODEL_NAME=gpt-4o-mini

LLM_BOOST_API_KEY=sk-your-key-here
LLM_BOOST_BASE_URL=https://api.openai.com/v1
LLM_BOOST_MODEL_NAME=gpt-4o

50 agents: ~1–3 minutes. Cost: ~$0.10–0.30 per run with gpt-4o-mini.

Other compatible providers

Provider	`LLM_BASE_URL`	Notes
Groq	`https://api.groq.com/openai/v1`	Fast, generous free tier
Together AI	`https://api.together.xyz/v1`	Good open-source models
LM Studio	`http://localhost:1234/v1`	Local GUI alternative to Ollama

Tuning the simulation

All of these are in simulate.py — no need to touch the backend.

"num_agents": 50,      # 10–30 = fast/rough, 50 = recommended, 100–200 = high confidence
"channel": "email",    # changes which social platforms personas prefer
"target_market": "...",# more specific → more varied, relevant personas

For deeper changes (conversion probabilities, persona archetypes, social post formats), see the advanced customization section below.

Worked example

The examples/codereview-ai/ folder contains a complete realistic simulation:

# See what output looks like — no backend needed:
python examples/codereview-ai/run_simulation.py --sample

It uses CodeReview AI ($29/mo, developer tooling) as the product and includes pre-computed realistic output showing 50 personas, 6 social posts, 5 objections, and 5 strategic insights.

Visual dashboard

Start the full stack (npm run dev) and open http://localhost:3000 to use the web UI instead of the terminal script.

After a simulation runs you'll get:

Metrics cards — conversion funnel at a glance
Adoption curve — purchase velocity over 72 hours
Social feed — all generated posts, filterable by platform and sentiment
Market Resistance — top objections with frequency
Strategic Insights — AI-generated recommendations
Persona interview panel — ask any generated persona any question in natural language

Advanced customization

Change conversion probabilities

backend/app/agents/launcher.py:

BASE_PROBABILITIES = {
    "ignored":   0.30,   # ← lower this to model high-awareness markets
    "viewed":    0.25,
    "clicked":   0.20,
    "read":      0.12,
    "purchased": 0.07,   # ← raise this for high-intent markets (e.g. enterprise inbound)
    "abandoned": 0.04,
    "shared":    0.02,
}

Change archetype purchase multipliers

ARCHETYPE_PURCHASE_MULTIPLIERS = {
    "early_adopter":   2.5,   # buys fast, forgives rough edges
    "pragmatist":      1.0,   # buys when clear ROI
    "conservative":    0.5,   # needs references, case studies
    "price_sensitive": 0.4,   # WTP is the main gate
    "skeptic":         0.2,   # rarely buys on first exposure
}

Change persona batch size

backend/app/agents/ethnographer.py:

BATCH_SIZE = 8    # increase to 15–20 if using GPT-4-class models

Add a social media platform

backend/app/agents/conversador.py:

PLATFORM_CONFIGS = {
    "twitter":      {"max_chars": 280,  "tone": "casual, punchy"},
    "reddit":       {"max_chars": 800,  "tone": "detailed, community-style"},
    "product_hunt": {"max_chars": 400,  "tone": "constructive, maker-friendly"},
    "app_store":    {"max_chars": 300,  "tone": "direct, star-rating style"},
    # Add your own platform here
}

Project structure

LaunchSimulation/
│
├── simulate.py                     ← RUN HERE — after creating input.txt
│
├── backend/
│   ├── .env.example                ← copy to .env, configure your LLM
│   └── app/
│       ├── agents/
│       │   ├── graph.py            # LangGraph pipeline
│       │   ├── researcher.py       # market research (DuckDuckGo + LLM)
│       │   ├── ethnographer.py     # persona generation
│       │   ├── populator.py        # ChromaDB indexing
│       │   ├── launcher.py         # interaction simulation + conversion funnel
│       │   ├── conversador.py      # social post generation
│       │   ├── chronicler.py       # metrics + strategic insights
│       │   └── interviewer.py      # RAG-powered persona chat
│       ├── api/                    # REST endpoints (simulate, results, interview)
│       ├── core/                   # config, LLM factory, database
│       ├── models/                 # Pydantic schemas + SQLAlchemy ORM
│       └── services/               # simulation orchestration + vector store
│
├── src/                            # Next.js frontend
│   ├── app/page.tsx                # product submission form
│   ├── app/simulation/[id]/page.tsx # results dashboard
│   └── components/                 # metrics, chart, social feed, interview panel
│
├── examples/
│   └── codereview-ai/              # worked example with sample output
│       ├── product_brief.md
│       ├── run_simulation.py
│       └── sample_output.json      # pre-computed — view without running backend
│
└── package.json                    # npm run dev · dev:backend · dev:frontend

API reference

If you prefer to call the API directly (curl, Postman, your own script):

Endpoint	Method	Description
`/api/simulate`	POST	Start a simulation
`/api/simulate/from-brief`	POST	Start from free-text product description
`/api/results/{id}`	GET	Full results
`/api/results/{id}/status`	GET	Lightweight status poll
`/api/results/{id}/personas`	GET	Generated personas only
`/api/results/{id}/posts`	GET	Social posts only
`/api/interview/{id}/agents`	GET	List interviewable personas
`/api/interview/{id}/{agent_id}`	POST	Ask a persona a question

Interactive docs: http://localhost:8000/docs

FAQ

How long does a simulation take? With Ollama (local, 7B model): 5–15 minutes for 50 agents depending on hardware. With Groq or OpenAI: 1–3 minutes.

How do I get more varied personas? Be more specific in target_market. "developers" produces similar personas. "junior backend developers in Southeast Asia earning under $30K" produces diverse, realistic ones.

Where is data stored?

Simulation results: backend/launchsim.db (SQLite)
Persona memories (for interviews): backend/chroma_db/

Delete both to reset everything.

Does this send my product description to the internet? Only if you use a cloud LLM (OpenAI, Groq, etc.). With Ollama, everything runs locally and nothing leaves your machine.

License

MIT — use it, fork it, ship it.

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Repository files navigation

LaunchSim

How to use it (3 steps)

Step 1 — Install

Step 2 — Describe your product

Step 3 — Run

What you get

Writing a good description

LLM options

Local — Ollama (free, private, no internet required)

Cloud — OpenAI

Other compatible providers

Tuning the simulation

Worked example

Visual dashboard

Advanced customization

Change conversion probabilities

Change archetype purchase multipliers

Change persona batch size

Add a social media platform

Project structure

API reference

FAQ

License

About

Uh oh!

Releases

Packages

Uh oh!

Contributors

Uh oh!

Languages

Name		Name	Last commit message	Last commit date
Latest commit History 7 Commits
backend		backend
examples/codereview-ai		examples/codereview-ai
frontend		frontend
.gitignore		.gitignore
README.md		README.md
compare_simulations.py		compare_simulations.py
input.txt		input.txt
input_example.txt		input_example.txt
package-lock.json		package-lock.json
package.json		package.json
simulate.py		simulate.py
test_quick.py		test_quick.py

Folders and files

Latest commit

History

Repository files navigation

LaunchSim

How to use it (3 steps)

Step 1 — Install

Step 2 — Describe your product

Step 3 — Run

What you get

Writing a good description

LLM options

Local — Ollama (free, private, no internet required)

Cloud — OpenAI

Other compatible providers

Tuning the simulation

Worked example

Visual dashboard

Advanced customization

Change conversion probabilities

Change archetype purchase multipliers

Change persona batch size

Add a social media platform

Project structure

API reference

FAQ

License

About

Topics

Resources

Uh oh!

Stars

Watchers

Forks

Releases

Packages 0

Uh oh!

Contributors

Uh oh!

Languages

Packages