Soon2026

SAMM Backend

This repo currently contains a minimal Next frontend scaffold and a Python SAMM backend foundation.

SAMM is built around a real LangGraph StateGraph: the CLI and API invoke/resume the graph directly, including native interrupt() checkpoints and Command(resume=...) resume.

Setup

python3 -m venv .venv
.venv/bin/python -m pip install -e '.[dev]'

Terminal Checks

.venv/bin/python -m samm demo --company "Celsius" --idea "hydration gummy for students and gym-goers" --mock
.venv/bin/python -m samm checkpoint --run-id <id> --decision approve
.venv/bin/python -m samm chat --mock
.venv/bin/python -m samm artifact --run-id <id>
.venv/bin/python -m samm events --run-id <id>
.venv/bin/python -m samm integrations --run-id <id>
.venv/bin/python -m pytest

For live OpenAI text generation, create a local .env with MOCK_MODE=false, OPENAI_API_KEY, and optionally OPENAI_MODEL. The research and asset upload integrations still use mock adapters until their real credentials are wired.

For live Composio + Exa research, also set COMPOSIO_API_KEY, EXA_API_KEY, COMPOSIO_EXA_AUTH_CONFIG_ID, COMPOSIO_EXA_CONNECTED_ACCOUNT_ID, and COMPOSIO_TOOLKIT_VERSION_EXA. The app can create the Composio Exa auth config and connected account at runtime if the IDs are missing, but storing the IDs in local .env avoids creating duplicate Composio connections.

Research depth is configurable with SAMM_EXA_SEARCH_TYPE, SAMM_EXA_BRAND_RESULTS, SAMM_EXA_MARKET_RESULTS, SAMM_EXA_FETCH_CONTENTS, SAMM_EXA_CONTENT_RESULTS, and SAMM_EXA_TEXT_MAX_CHARACTERS. With Composio's current Exa toolkit, SAMM uses staged EXA_SEARCH calls plus EXA_GET_CONTENTS_ACTION for deeper source context.

API

.venv/bin/python -m uvicorn samm.api:app --reload

The API exposes /health, run creation, conversational messages, checkpoint resume, iteration, artifact inspection, event timelines, and integration events.

New runs first pause at an idea_calibration checkpoint so the orchestrator can ask how to explore the concept before starting Brand Intelligence and Market Opportunity research. Submit run_it to infer defaults, or pass a selected direction/free-text message to store as intake context before research.