SalesBuff — Pre-call intelligence + live coaching

Speak your sales scenario and SalesBuff helps before and during the call: Pre-call turns a one-liner into citation-grounded Action tips and a Fact dossier; On-fly streams real-time coaching moves as the conversation unfolds.

SalesBuff resolves the real companies and people involved, researches them across the web (and court records for pre-call), and produces citation-grounded output — no chat, no fluff.

---

What it does

Pre-call (before the meeting)

You give it something like:

"Meeting Dr. Lee Schwamm at Yale New Haven Health tomorrow. We're expanding our AI clinical-documentation rollout to more physicians."

…and it returns:

Tab	What you get
Actions	Conversation moves grouped by call flow — opening move, pain hypothesis, differentiation, proof point, next step. Each is something to say / ask / show / avoid / verify, with a talk-track and sources.
Facts	An evidence dossier grouped into domain categories (priorities, pain signals, stakeholders, legal/regulatory, open questions), every finding backed by a source URL.

Both come from one shared research pass, so the two tabs always agree on the evidence.

On-fly (during the call)

Start a live session with optional pre-call context, speak into the mic, and get verb-first coaching tips as the call progresses:

• Conversation memory — compacts older transcript into a structured summary while keeping a raw tail of recent speech.
• Bootstrap + reactive research — enriches thin context at session start; mid-call Tavily lookups (quick or deep multi-query) when the coach detects gaps (competitor, objection, ROI proof, news).
• Stage-aware tips — icebreaker → discovery → differentiation → close, with dedup and quality filters so cards stay actionable, not generic.
• Session logs — full JSON event trace dumped when the session ends (for tuning and debugging).

---

Repository layout

SalesBuff-Pre-Call/                ← this repo (git root)
├── README.md                      ← you are here
├── CONTRIBUTING.md                ← how to contribute / push / deploy
├── CONTEXT.md                     ← domain glossary (pre-call + on-fly)
├── render.yaml                    ← Render blueprint for the backend
└── SalesBuff/
    ├── README.md                  ← backend guide (Python / FastAPI)
    ├── salesbuff/                 ← the Python package (API + pipeline + on-fly)
    └── SalesBiff-Frontend/
        ├── README.md              ← frontend guide (TanStack Start)
        └── ...

• Backend — Python + FastAPI. Core (LLM + search clients) shared by Pre-call (resolve → research → brief) and On-fly (live coaching). See SalesBuff/README.md.
• Frontend — TanStack Start + Vite + React + Tailwind. Pre-call brief + On-fly live coach in one app; five build-time UI themes (VITE_THEME). See SalesBuff/SalesBiff-Frontend/README.md.
• Architecture deep-dive — SalesBuff/ARCHITECTURE.md.

---

Python package — salesbuff on PyPI

The engine is published as a standalone Python package so you can use it directly in your own code — no server needed.

pip install salesbuff                        # shared core
pip install "salesbuff[precall]"             # + pre-call due-diligence feature
pip install "salesbuff[onfly]"               # + live coaching (Core only — no precall deps)
pip install "salesbuff[precall,api]"         # + FastAPI host (run salesbuff-serve)
pip install "salesbuff[all]"                 # everything

Pre-call SDK:

from salesbuff import SalesBuff

async with SalesBuff(openai_api_key="sk-...", tavily_api_key="tvly-...") as sb:
    result = await sb.research(
        "Meeting the VP of Ops at Acme Health next week..."
    )
    print(result.brief)   # Actions coaching brief
    print(result.facts)   # Evidence dossier

One-shot sync helper for scripts/notebooks:

from salesbuff import research_once

result = research_once("...", openai_api_key="sk-...", tavily_api_key="tvly-...")

Host it yourself from the terminal (no code needed):

pip install "salesbuff[precall,api]"
# set OPENAI_API_KEY + TAVILY_API_KEY in your env, then:
salesbuff-serve --port 8000

On-fly live coaching is exposed at /onfly/* when the API is running (see backend README for routes).

📦 pypi.org/project/salesbuff · Current version: 0.2.0

---

Quickstart (local)

Run the two services in separate terminals.

1) Backend (needs OpenAI + Tavily keys)

cd SalesBuff
pip install -r salesbuff/requirements.txt
cp salesbuff/.env.example salesbuff/.env   # then fill in your keys
uvicorn salesbuff.api:app --port 8000 --reload

2) Frontend

cd SalesBuff/SalesBiff-Frontend
npm install
npm run dev            # talks to http://127.0.0.1:8000 by default

Open the printed local URL. Use Pre-call to run due diligence, or switch to On-fly for live coaching (mic + optional pre-call context).

No keys? The UI also lets a user paste their own OpenAI + Tavily keys (kept in the browser tab only) to run without touching the shared quota.

---

How the pieces talk

Browser ──► Frontend server fns ──► Backend API ──► OpenAI + Tavily + CourtListener
                (SALESBUFF_API_URL)
                     ├── /research/*   pre-call jobs
                     └── /onfly/*       live coaching sessions

The browser never calls the backend directly — TanStack server functions proxy to it, so the backend URL and any keys stay off the client.

---

Deployment

Part	Host	Notes
Backend	Render	Web service, Root Directory = SalesBuff. Blueprint in render.yaml. Long jobs + in-memory state → keep 1 instance.
Frontend	Vercel	Root Directory = SalesBuff/SalesBiff-Frontend, Nitro vercel preset (set in vite.config.ts). Set SALESBUFF_API_URL to the Render URL. Set VITE_THEME per project (sunrise, prism, horizon, folio, or ember).

Full steps and env vars are in CONTRIBUTING.md and each sub-README.

---

Security

• Never commit secrets. .env files are git-ignored — use .env.example as the template and set real values in Render / Vercel.
• If a key is ever exposed, rotate it immediately.