What makes it different: the CRM doesn't just store your data, it actively works your pipeline. AI agents autonomously discover and qualify leads by crawling the web. A built-in voice AI dialer (AWS Chime + Azure) handles outbound calls. An AI insights engine runs neural health checks across your revenue, pipeline, and team performance in real time.
The full GTM lifecycle is covered end-to-end: Campaigns, LeadGen, Accounts, Lists, Contacts, Leads, Opportunities, Quotes, Contracts, Invoices, and Projects — all linked. We also ship a visual workflow builder, a Mermaid diagram editor, multi-channel outreach (email, voice, SMS), and a built-in University module with certification paths and a RevOps simulator.
Tech stack: Next.js 16, React 19, Prisma, MongoDB, Vercel AI SDK (OpenAI, Anthropic, Azure, Google, Mistral), AWS Chime, React Email, Resend, shadcn/ui, Tremor, Framer Motion.
Live demo at crm.basalthq.com (free to enroll). We're shipping fast and would love feedback from the HN community.
@BasaltHQ - https://x.com/BasaltHQ
Why an OS, not a point solution: Most AI-in-enterprise approaches bolt a model onto one workflow. We built a shared execution environment where agents have defined skills, scoped permissions, and structured communication. 21 agents, 55 registered skills, 800+ API endpoints. Agents operate within a skill chainer that composes multi-step executions across agent boundaries.
Constitutional governance — hardcoded, not prompted: ODE enforces 10 constitutional laws. These aren't system prompts. They're hardcoded constraints checked on every agent action at the application layer. No agent can authorize its own financial exposure. No single agent can override multi-tenant data isolation. Every decision must produce a verifiable audit record.
3-head decision protocol: High-risk actions require sign-off from three executive agents evaluating orthogonal dimensions: operational impact, constitutional compliance, and financial exposure. All three must approve. Inspired less by AI safety literature and more by nuclear launch protocols — independent verification across different concern axes.
Crypto-backed provenance: Every agent decision produces a cryptographically verifiable audit trail. In regulated industries, "the AI decided" isn't acceptable.
Agent-to-agent commerce: External AI agents can discover ODE's capabilities via .well-known/agents.json, then hire and pay ODE agents through an A2A protocol. Not agent orchestration within a system — agent commerce across organizational boundaries.
What's open source: 15,000 enterprise use cases on HuggingFace: https://huggingface.co/datasets/LlewellynSystems/ode-enterpr...
Stack: Next.js 15, Express.js 5, Supabase PostgreSQL, Prisma ORM.
Honest bottleneck: Latency on 3-head consensus under load — p99 gets uncomfortable on complex actions. Agent commerce dispute resolution between organizations is a hard open problem.
Happy to go deep on any of this.
