autonomous voice buyer that calls, compares, and haggles — with receipts.
built for ElevenLabs Hack-Nation · The Negotiator
live product · real outbound calls · verified leverage only · never invents competitor prices
scout-demo.mp4
| Evidence | Conversation | Video | Audio | What it shows |
|---|---|---|---|---|
| Product walkthrough | — | Playing above | — | Confirmed property brief → discovery → parallel negotiation → evidence-backed recommendation |
| Negotiation call 1 | conv_5201kxx76rwvebfb1s430ve2nm5h |
Open inline player | Listen | Hindi-first hostel availability, scope, and price discussion |
| Negotiation call 2 | conv_9401kxx51dk6fxgveay6375kr57n |
Open inline player | Listen | Bounded, truthful India-hostel negotiation and clean close |
▶ Play video and listen to the negotiation
negotiation-call-1.mp4
▶ Play video and listen to the negotiation
negotiation-call-2.mp4
people shopping for high-friction local services (movers, hostels, contractors) waste hours calling vendors one-by-one. quotes are inconsistent. leverage dies between calls.
Scout turns one confirmed brief into:
- discovery (OpenStreetMap)
- parallel voice calls (ElevenLabs → Twilio)
- itemized, comparable quotes
- haggling that may only cite verified competing bids
- a ranked recommendation with transcript evidence
hard rule: Scout never pays, signs, or binds the customer. AI discloses when asked.
first vertical: moving. live multilingual path: India hostel / PG (Hindi-first + Indian English).
ElevenLabs · Twilio · Next.js · Node orchestrator · OpenStreetMap · Tavily · Gemini / Claude (via ElevenLabs) · OpenAI helpers · pnpm monorepo · simulated market · verified-leverage strategy engine
honesty is enforced in code — the voice LLM speaks; the orchestrator decides tactics and which numbers are allowed.
apps/
web/ intake, live activity, comparison UI (Vercel)
orchestrator/ call lifecycle, strategy, ranking, audit
voice-agents/ ElevenLabs intake + negotiator prompts/tools
simulated-market/ deterministic seller styles (no phone credits)
packages/
contracts/ RequirementSpec · CallOutcome · Quote · Recommendation
vertical-config/ fees, red flags, negotiation levers
evals/ golden-call fixtures
media/demo/ demo video · bargaining audio · source zip
docs/ architecture, submission copy, Notebook LM prompt
START-HERE.md how to read this repo / the zip
- this README (demo links above)
START-HERE.md— reading order for the zip / new contributorsdocs/ARCHITECTURE.md— system boundariesdocs/negotiation-engine.md— verified leveragedocs/voice-and-multilingual.md— Hindi / accent pathapps/voice-agents/src/agents/negotiator.md— what the agent is allowed to sayapps/orchestrator/src/negotiation/strategy-engine.js— next move brain
Hack-Nation form copy: docs/hack-nation-submission.md
- one confirmed requirement spec reused across every call
- ≥3 distinct counterparty / negotiation styles
- itemized, comparable quote outcomes
- at least one genuine, evidence-based price/term change
- ranked recommendation linked to transcript evidence
- honest AI disclosure · no fabricated bids · no authority to commit funds
pnpm install
pnpm --filter @scout/orchestrator serve
pnpm --filter @scout/web devopen http://localhost:3000 — or jump to /moving / /real-estate
outbound calls stay disabled until OUTBOUND_CALLS_ENABLED=true and keys are set (see .env.example).
no-credit smoke path:
node apps/simulated-market/src/simulator.smoke.js
node apps/orchestrator/src/moving/moving-workflow.smoke.jslive ElevenLabs tool boundary: docs/AGENT-TOOLS.md
- frontend: https://scout-dusky-six.vercel.app
- source zip for judges:
media/demo/scout-hacknation-source.zip
| challenge | The Negotiator — voice agents that call, compare, and haggle |
| team id | HN-1602 |
| github | nothariharan/scout |
| live | scout-dusky-six.vercel.app |