OpenTeleCRM

The self-hosted, drop-in replacement for TeleCRM. A telecalling-first sales CRM for teams that live on the phone — leads, dialer, WhatsApp, automation — with your data on your hardware and an API your existing integrations already speak.

Table of contents

• What is OpenTeleCRM?

• See it in action

• Features

• Architecture

• Quick start

• TeleCRM parity

• Roadmap

• Documentation

• Tunnel mode

• Contributing

• License

Related MCP server: Outlet das Cores CRM

What is OpenTeleCRM?

OpenTeleCRM is a 1:1 open-source clone of TeleCRM — the telecalling-first sales CRM used by thousands of Indian sales teams — rebuilt clean-room against the documented API surface, self-hosted natively (no Docker), and multi-tenant from line one.

It answers the three questions every telecalling operation asks:

The parity promise is compatible surface, not bug-compatible behavior: where TeleCRM silently drops async writes, caps tokens at 3, or expires MCP tokens in 30 days with no renewal, OpenTeleCRM deliberately fixes it (see docs/PARITY.md § Divergences).

See it in action

Agent desk (dark theme) — dashboard, callbacks queue, automation rules, all rendered against the seeded demo workspace:

Features

Architecture

A pnpm + Turborepo monorepo, ESM-only, Node ≥ 22. One Postgres database, one tenant-scoping discipline:

flowchart LR
    WEB[Next.js agent desk<br/>apps/web · :3007] --> API[NestJS API<br/>services/api · :3005]
    MOB[Kotlin mobile app<br/>apps/mobile] --> API
    MCPC[MCP clients] -->|JSON-RPC /mcp| MCP[MCP server<br/>services/mcp · :3100]
    API --> DB[(PostgreSQL 16/17<br/>28 RLS-FORCE tenant tables)]
    MCP --> DB
    API --> BRIDGE[WhatsApp bridge<br/>services/whatsapp-bridge · :3098]
    BRIDGE -.->|Baileys 7.x| WA[WhatsApp]
    API --> ARI[Asterisk 21 LTS · ARI<br/>infra/asterisk · :8088 loopback]
    ARI -.->|SIP| PSTN[PSTN / SIP trunk]

Every request resolves a token → tenant, then reads through withTenant(eid) — a transaction that sets app.enterprise_id so RLS returns only that tenant's rows. Missing tenant context returns zero rows, not a leak.

Full detail — C4 diagrams, data flow, ports, future containers: docs/ARCHITECTURE.md

MCP client setup

The MCP server (services/mcp) exposes 13 TeleCRM-parity tools over Streamable HTTP (POST /mcp, loopback bind by default). In production it refuses to boot without MCP_ENTERPRISE_ID and either MCP_BEARER_TOKEN or an explicit MCP_ALLOW_UNAUTHENTICATED=true; set MCP_HOST=0.0.0.0 only when you also set a bearer token (a public bind without one is rejected).

Server env (services/mcp):

MCP_PORT=3100
MCP_ENTERPRISE_ID=<tenant-uuid>
MCP_BEARER_TOKEN=<long-random-token>   # recommended
MCP_HOST=127.0.0.1                     # default; loopback only

Clients that support auth-header streamable HTTP configs: Claude Desktop, Cursor, VS Code MCP, MCP Inspector, Continue — all use the same shape:

Claude Desktop — claude_desktop_config.json:

{
  "mcpServers": {
    "opentelecrm": {
      "type": "http",
      "url": "http://127.0.0.1:3100/mcp",
      "headers": { "Authorization": "Bearer <MCP_BEARER_TOKEN>" }
    }
  }
}

Cursor — .cursor/mcp.json:

{
  "mcpServers": {
    "opentelecrm": {
      "type": "http",
      "url": "http://127.0.0.1:3100/mcp",
      "headers": { "Authorization": "Bearer <MCP_BEARER_TOKEN>" }
    }
  }
}

Point url at the tunnel/ingress host when the server is remote, and keep the bearer token out of anything you commit (.gitignore already covers *.mcp).

Quick start

Prerequisites: Debian/Ubuntu (or any Linux), Node.js ≥ 22 (corepack pnpm), PostgreSQL 16 or 17, ~10 minutes.

git clone https://github.com/iamsrishanth/OpenTeleCRM.git
cd OpenTeleCRM

make setup      # provision system deps + pnpm install + db-init + db-migrate + db-seed
make dev        # API on :3005  (web desk: pnpm --filter @opentelecrm/web dev → :3000)
pnpm test       # 151 tests across API / MCP / whatsapp / telephony

make setup is a one-shot: it provisions native deps, installs JS deps, creates the DB role, runs all 8 migrations, and seeds a demo workspace (5,000 leads, 3 users, 2 pipelines). Step-by-step targets (make provision, make install, make db-init, make db-migrate, make db-seed) exist if you'd rather run them separately.

Bruno collection (wire-compatibility proof):

cd collections/opentelecrm
bash ../../scripts/bruno-bootstrap-jwt.sh   # inject a dev JWT
npx @usebruno/cli run --env local -r .       # 27 requests, 26 assertions

TeleCRM parity

Full matrix with per-feature test IDs: docs/PARITY.md

Roadmap

Shipped: P0–P4 (foundation → core CRM → WhatsApp → telephony → automation), P4b (web desk, visual builder, sequences, quota metering, live Asterisk/WhatsApp wiring), P8 mobile (ahead of plan, as Kotlin native).

Next up: P5 lead capture (connector SDK, persistent ingest log, webhook / CSV / missed-call / email / FB Lead Ads connectors). Then P6 analytics, P7 AI & voice, P9 admin/migration/SaaS, P10 hardening & launch.

Per-phase scope + exit criteria: docs/ROADMAP.md

Documentation

Networking & tunnel mode

The web desk resolves its API base at runtime from the browser's own origin (apps/web/src/lib/config.ts getApiBase()), so it works from every surface without a build-time flag:

The Cloudflare tunnel (optional — demos / remote teams) splits web and API onto two hostnames; both are managed by scripts/tunnel.py through the Cloudflare API:

make tunnel      # ensure DNS CNAMEs + ingress: crm.srishanth.com -> :3007 (web),
                 # api.srishanth.com -> :3005 (API); write apps/web/.env.local
make untunnel    # revert to local mode

Tunnel origins, zone, and all Cloudflare credentials live only in gitignored files (.env, apps/web/.env.local, /etc/cloudflared/token) — never committed, never printed. Because the API base is derived at runtime, no next build time flag or baked origin is involved — the same bundle serves every surface. While the tunnel is up the API is publicly reachable, so never run it with production credentials.

Cross-platform supervision

The web + API servers are supervised identically on Linux and macOS through one portable launcher pair (infra/launchers/launch-api.sh, infra/launchers/launch-web.sh — resolve node dynamically, source .env, no watch flag):

• Linux (systemd): infra/systemd/opentelecrm-{api,web}.service (sudo systemctl enable --now opentelecrm-api opentelecrm-web).

• macOS (launchd): infra/macos/com.opentelecrm.{api,web}.plist (launchctl load ~/Library/LaunchAgents/…).

Full runbook + portability matrix: infra/macos/README.md.

Contributing

Contributions are welcome — this is a young project and the roadmap is long.

1. Fork → branch → PR (conventional commits).

2. The gate is green or the PR doesn't land: pnpm test (151), make typecheck, pnpm lint (Biome for services/packages; eslint for apps/web).

3. Docs update in the same commit as the feature — test counts, tables, ports, and statuses in README/PARITY/ROADMAP must not drift (see docs/README.md conventions).

4. No Docker, ever — native provisioners only: scripts/provision/debian.sh on Linux, infra/macos/provision-brew.sh on macOS (ADR-0001).

5. No secrets — .env*, tunnel hostnames, *.mcp tokens, and the mobile release keystore are gitignored; a PR that commits any of them will be rejected.

License

AGPL-3.0 — see LICENSE and docs/LICENSES.md.

OpenTeleCRM is not affiliated with TeleCRM. It is an independent, clean-room implementation of TeleCRM's documented API surface for self-hosted use. TeleCRM is a trademark of its respective owner.