___  _________  __  _________
  / __/ / ___/ _ \/ / / /_  __/
 _\ \  / /__/ // / /_/ / / /
/___/  \___/\____/\____/ /_/   intelligence collector

A lightweight, containerized agentic intelligence collector. Give it a mission in plain English; Scout reasons about where the intelligence lives, sends a team of agents to collect it, maps what it finds into a graph, and hands back a sourced report.

• Run it free, or bring your own model. Scout is model-agnostic. Start on a free no-credit-card tier (Groq, Google Gemini, or NVIDIA NIM) in ~2 minutes, point it at a frontier key (Claude, GPT) for the richest results, or run it fully offline against a local model.
• 33 specialized OSINT tools, auto-detected. Beyond web search and fetch, Scout ships a pluggable library spanning accounts, email, phone, social, domain/recon, sanctions screening, crypto, news, and more. It probes your environment at startup and hands agents only the tools that are usable - and tells you which relevant ones you could enable.
• Every source is fair game. Web, APIs, your local documents, and gated or offline knowledge - when an agent needs credentials or first-hand input, it pauses and asks you in the terminal.

A working, self-hostable foundation for agentic data collection - not a managed product. Pull it, run it in Docker, point it at a goal, and watch agents plan, search, read, ask you for help when they're blocked, and assemble a sourced report. Early and single-operator by design, but real.

Get started free (about 2 minutes)

No paid account, no credit card. Scout runs on free model tiers. You need Docker and git.

1. Get a free API key (pick one):

• Groq (recommended - fastest): open console.groq.com/keys, sign in with Google or email, click Create API Key, and copy the key (it starts with gsk_).
• Google Gemini: open aistudio.google.com/apikey, sign in, click Create API key, and copy it.
• NVIDIA NIM: open build.nvidia.com, make a free developer account (no card), and copy your nvapi- key. Gives you DeepSeek and 80+ other models.

2. Get the code and config file:

git clone https://github.com/tllongdev/scout.git
cd scout
cp .env.example .env

3. Paste your key into .env. Open .env in any editor and set the two lines for the provider you chose:

# If you used Groq:
SCOUT_MODEL=groq/llama-3.3-70b-versatile
GROQ_API_KEY=gsk_paste_your_key_here

# Or if you used Gemini instead:
# SCOUT_MODEL=gemini/gemini-2.5-flash
# GEMINI_API_KEY=paste_your_key_here

# Or if you used NVIDIA instead (DeepSeek with tool calling):
# SCOUT_MODEL=nvidia_nim/deepseek-ai/deepseek-v3.2-exp
# NVIDIA_NIM_API_KEY=nvapi-paste_your_key_here

4. Run your first mission:

./run.sh "Investigate Acme Robotics and its key partners"

That's it. Results land in ./output/: a findings.md report and an interactive graph.html you can open in your browser.

No key yet and just want to watch it run? Use SCOUT_MODEL=mock ./run.sh "..."

• it runs the whole pipeline with scripted, illustrative output (no key, no cost).

How it works

mission brief
     │
     ▼
┌──────────┐   decomposes the goal into focused collection tasks
│ Planner  │   (reasons about web, APIs, local docs, and gated/offline sources)
└────┬─────┘
     │        ┌─ auto-detects usable OSINT tools for this run, and
     ▼        │  flags relevant ones you haven't enabled yet
┌──────────┐ ─┘
│Collectors│   one agent per task, each with:
│          │   core tools:  web_search · web_fetch · local_files · ask_human
│          │   + 33 specialized OSINT tools (accounts, email, phone, social,
│          │     recon, sanctions, crypto, news, geo...) - whichever are usable
│          │   + records entities, relationships, observations as it goes
└────┬─────┘
     │
     ▼
┌──────────┐   draws sourced conclusions over the collected graph
│Synthesis │
└────┬─────┘
     │
     ▼
  output/  →  findings.md · graph.html · graph.json · raw_documents.json

The specialized tools are pluggable and discovered at startup (see Tool library): Scout only hands an agent the tools that are actually usable given your keys, packages, and binaries, and lists the relevant ones you could turn on.

Nothing is off the table as a source - including material that isn't on the internet. When an agent needs credentials, access to a gated system, or first-hand knowledge, it pauses and asks you right in the terminal.

Tool library

Beyond web search and fetch, Scout ships a pluggable library of specialized OSINT tools. At mission start it auto-detects which tools are usable - based on the keys, packages, and binaries you have - and hands only those to the agents (same idea as model discovery). Run scout tools to see live status.

Keyless tools work out of the box; the rest light up when you add a key, install an extra (pip install "scout[osint]"), or point Scout at a cloned repo.

If a mission would clearly benefit from a tool you haven't enabled, Scout tells you - at the start of the run and in the report it lists the relevant unconfigured tools and the exact step to turn each one on.

Looking for more tools to add? scout discover queries GitHub for the top-starred OSINT repos and flags which Scout already integrates, so you can spot strong new candidates to wrap into the registry (scout discover phone to focus the search). The agents themselves work the mission by every legitimate means (reasoning, web search/fetch, reading your files, asking you), and reach for a specialized tool whenever the task is something that tool is built for.

Tool	Category	What it's for	Needs
airplanes.live	aviation	Live ADS-B flight tracking (incl. military)	nothing (free)
web-check	web recon	DNS, SSL, headers, WHOIS, ports for a domain	nothing (free)
Reddit (PullPush)	social	User history + search, incl. removed content	nothing (free)
Data-broker opt-out	data hygiene	Removal links for people-search brokers	nothing (free)
GDELT News	news	Near-real-time global news/event search	nothing (free)
Telegram OSINT	social	Read public Telegram channel posts	nothing (free)
Crypto Wallet Trace	crypto	BTC/ETH balance + activity via explorers	nothing (free)
NVD CVE Lookup	cyber	CVEs by id or product keyword (CVSS)	nothing (free)
Scrapling	web recon	Anti-bot / stealth scraping	pip install scrapling
GeoCLIP	geolocation	Open-source image → GPS, offline	pip install geoclip
image-matching-webui	imagery	Feature-match two images to corroborate scenes	pip install imcui
Presidio	data hygiene	Redact PII from collected text	pip install presidio-analyzer
BBOT ⚠	recon	Recursive attack-surface / subdomain recon	pip install bbot
GHunt	accounts	Google account footprint from an email	pip install ghunt + login
maigret	accounts	Deep username dossier (3,000+ sites, profile parsing)	pip install maigret
Blackbird	accounts	Fast username/email search (WhatsMyName, 600+ sites)	clone repo, set SCOUT_BLACKBIRD_PATH
holehe	email	Which of 120+ sites an email is registered on	pip install holehe
theHarvester	email	Emails/names/subdomains for a domain	pip install theHarvester
PhoneInfoga	phone	Phone-number validation + footprint	phoneinfoga binary on PATH
dnstwist	web recon	Typosquat / lookalike / phishing domains	pip install dnstwist
Photon	web recon	Crawl a site for emails/handles/links/secrets	clone repo, set SCOUT_PHOTON_PATH
Instaloader	social	Public Instagram profile metadata	pip install instaloader
Toutatis ⚠	social	Deep Instagram extraction (email/phone)	pip install toutatis + TOUTATIS_SESSION_ID
subfinder	web recon	Fast passive subdomain enumeration	subfinder binary on PATH
httpx	web recon	Probe hosts for liveness/title/tech	ProjectDiscovery httpx binary (set SCOUT_HTTPX_PATH if it clashes with python httpx[cli])
SpiderFoot ⚠	recon	All-in-one OSINT engine (200+ modules)	clone repo, set SCOUT_SPIDERFOOT_PATH
geowifi	geolocation	Geolocate a WiFi BSSID/SSID	clone repo, set SCOUT_GEOWIFI_PATH
Robin ⚠	dark web	LLM-driven dark-web search (needs Tor)	robin on PATH
OpenSanctions	sanctions	Screen names vs OFAC SDN / PEP / watchlists	OPENSANCTIONS_API_KEY (free for journalists/NGOs)
Grayhat Warfare ⚠	exposure	Search exposed S3/Azure/GCS buckets	GRAYHAT_API_KEY
FaceCheck.ID ⚠	facial recognition	Reverse face search (paid)	FACECHECK_API_TOKEN
GeoSpy ⚠	geolocation	AI photo geolocation (commercial)	GEOSPY_API_KEY
MarineTraffic / Kpler	maritime	Vessel tracking by MMSI/IMO (commercial)	MARINETRAFFIC_API_KEY

⚠ = carries legal/ethical/reputational weight (surveillance, facial recognition, exposed data, dark web). Use responsibly, lawfully, and within each service's terms. Control what's active with SCOUT_TOOLS / SCOUT_DISABLE_TOOLS (see .env.example).

Bring your own model

Scout ships no credentials. You choose the model and supply your own key, so it works with whatever you already pay for - or a model running on your own hardware. Set this in .env:

For best results, use a frontier model. Scout's planning, multi-step tool use, and synthesis get sharply better with model quality. We recommend Claude Sonnet 4.6 for the best balance of depth, speed, and cost, or Claude Opus 4.8 for maximum extraction depth. The free Groq and Gemini tiers are great for trying Scout out - just expect richer graphs from a frontier model.

You have...	SCOUT_MODEL	Also set
Nothing - want a free real model	groq/llama-3.3-70b-versatile	GROQ_API_KEY (free, no card)
A free Gemini key	gemini/gemini-2.5-flash	GEMINI_API_KEY (free tier)
A free NVIDIA key	nvidia_nim/deepseek-ai/deepseek-v3.2-exp	NVIDIA_NIM_API_KEY (free, no card)
A Claude / Anthropic key	anthropic/claude-sonnet-4-6 (or anthropic/claude-opus-4-8)	ANTHROPIC_API_KEY
An OpenAI key	openai/gpt-5.5	OPENAI_API_KEY
Qwen on your home server (Ollama)	ollama/qwen2.5:14b	SCOUT_API_BASE=http://host.docker.internal:11434
Any OpenAI-compatible server (vLLM, LM Studio, TGI)	openai/<name>	SCOUT_API_BASE=http://host.docker.internal:8000/v1
Just want to see it run offline	mock	nothing (no key, scripted output)

Free tiers (Groq, Gemini, NVIDIA NIM) are rate-limited but plenty for a test mission - see Get started free above for the 2-minute setup. For heavy runs, use a paid or local model. The mock option runs the full pipeline with scripted output and needs no key at all.

Not sure which model string to use?

Scout can list the models your credentials can actually reach - it asks each provider's own API (and your local Ollama box), so you get real, current options rather than a guess:

docker compose run --rm scout models

And if you run a mission without SCOUT_MODEL set, Scout discovers your available models and lets you pick one interactively for that run.

Running missions

Already set up (see Get started free)? A few more ways to run:

# Give agents offline/private documents to read (optional)
mkdir -p sources
cp ~/some-dossier.txt sources/

# Run a mission
./run.sh "Map the leadership and funding relationships of the top 5 AI safety nonprofits"

# Or without the helper script
docker compose run --rm scout "Your mission here"

Results land in ./output/mission-<date>-<id>/:

• findings.md - the report: summary, ranked findings, entities, relationships
• graph.html - interactive intelligence graph (open in a browser)
• graph.json - the graph as node-link data
• raw_documents.json - every source collected, verbatim, for audit

Tuning

Set these in .env:

• SCOUT_MAX_TASKS - how many collection tasks the planner may create (default 6)
• SCOUT_MAX_STEPS - max tool calls per collector agent (default 12)
• SCOUT_TEMPERATURE - model temperature (default 0.4)

Local development (without Docker)

uv venv && source .venv/bin/activate
uv pip install -e .
cp .env.example .env   # add your model + key
scout "Your mission here"

Notes

• This is an early, single-operator tool, not a hardened production service. Be a good citizen: respect site terms, rate limits, and the law.
• The ask_human flow makes Scout safe around credentials - it never invents secrets; it asks, and you can mask the input.
• Graph quality scales with model quality. A frontier model produces noticeably richer entity/edge extraction than a small local one.