A multi-tenant agentic platform for personal & organizational context management, built around the Personal Context Management Service (PCMS) — a graph-backed memory layer that lets autonomous agents reason over a tenant's email, calendar, tasks, docs, and meetings while preserving strict tenant isolation.
The platform is composed of:
- PCMS (
services/pcms/) — FastAPI service deployed on Cloud Run; owns Memgraph (graph), Qdrant (vectors), Redis (queues), and tenant lifecycle. - Hermes (
agents/hermes/) — long-running agent framework with a reinforcement-learning loop that promotes successful trajectories from the graph into agent prompts. - OpenClaw (
agents/openclaw/) — desktop / browser agent runtime that shares the same Memgraph backend via an MCP server, giving every agent the same memory. - Nango (
nango/) — managed sync layer that pulls Gmail, Google Calendar, Microsoft Teams, Asana, JIRA, and Confluence into PCMS. - Dashboard (
dashboard/) — Firebase-hosted control plane for tenants, agents, and the chat UI.
┌─────────────────────────────┐
│ Firebase Dashboard │
│ (firebase_public/) │
└────────────┬────────────────┘
│ HTTPS / JWT
▼
┌───────────────────────────────────────────────┐
│ PCMS (Cloud Run, services/pcms/) │
│ ─ tenants, auth, memories, intel, webhooks │
└─┬──────────┬───────────┬───────────┬──────────┘
│ │ │ │
▼ ▼ ▼ ▼
Memgraph Qdrant Redis Nango Cloud
(graph) (vectors) (queues) (Gmail/GCal/Teams/
Asana/JIRA/...)
▲ ▲
│ Bolt │ HTTP
│ │
┌──────┴──────┐ │
│ Hermes RL │ │
│ agent loop │ │ ┌────────────────────────┐
│ (rl/, env) │ └───────┤ OpenClaw runtime │
└─────────────┘ │ + memgraph MCP server │
└────────────────────────┘
Both Hermes and OpenClaw write trajectories into the same Memgraph instance
through the memgraph-memory-sync hooks; the RL crystallizer promotes
high-reward trajectories into agent prompts on the next run.
Prereqs: a Mac or Linux box with Docker, Node.js 20+, Python 3.11+, gcloud CLI, firebase CLI, gh CLI.
git clone https://github.com/morganjppeach/peach-pilot-platform.git
cd peach-pilot-platform
cp .env.example .env # edit and fill in real values
bash scripts/setup-all.sh # runs all of the per-component scriptssetup-all.sh does the following, skipping steps whose state already exists:
- checks prerequisites
- brings up local Memgraph + Redis via
infrastructure/docker-compose.yml - (optional) creates GCP project resources via
scripts/setup-gcp.sh - installs and configures Hermes (
scripts/setup-hermes.sh) - installs and configures OpenClaw (
scripts/setup-openclaw.sh) - configures Nango integrations (
scripts/setup-nango.sh) - deploys the Firebase dashboard (
scripts/setup-firebase.sh) - provisions a test tenant (
scripts/create-tenant.sh test-tenant) - runs
ops/run-health-checks.sh - prints the tenant API key, dashboard URL, and PCMS URL
Detailed setup is in SETUP.md. Other key docs:
- Product
docs/PRD_AGENT_SPRINTS_v4.md— current PRDdocs/AGENT_HIERARCHY_v2.md— agent rolesdocs/PRD_ROADMAP_REVIEW_v2.md— roadmapdocs/BUSINESS_PLAN_v2.md
- Engineering
- Operations
docs/SEAN_RUNBOOK.md— on-call runbookops/system-health-tests.md— health checks
pip install -e services/pcms
pytest tests/ -qThe tests/rl/ suite requires Memgraph running on localhost:7687
(spin it up with docker compose -f infrastructure/docker-compose.yml up -d).
peach-pilot-platform/
├── services/pcms/ FastAPI multi-tenant memory service
├── agents/
│ ├── hermes/ Hermes RL system + hooks + Memgraph env
│ └── openclaw/ OpenClaw hooks + memgraph MCP server
├── infrastructure/ Dockerfiles, docker-compose, cloudbuild.yaml
├── nango/ Nango integrations (Gmail / GCal / Teams / ...)
├── dashboard/ Firebase static dashboard
├── docs/ PRDs, business plan, runbooks
├── ops/ Health checks, runbook, alert configs
├── tests/ unit / integration / e2e / RL
└── scripts/ setup-all.sh + per-component installers
Proprietary — Morgan & Peach. Internal use only unless otherwise noted.