DinoClaw
A conversational AI agent that runs your business — not just a chatbot bolted on top.
Every "AI for SMB" play has been a chatbot bolted onto a generic SaaS. Either the AI doesn't have access to real business data or the underlying system can't actually take action. Owner-operators end up with another tab to ignore.
DinoClaw flips the model: the agent is the primary surface. Behind it sits a full ERP — CRM pipeline, jobs and crews, double-entry accounting, multi-warehouse inventory with FIFO costing, billing, HR. The agent has 127 typed tool operations against this stack and can take real actions on it. Skill-based personality keeps the conversation warm and consultative; underneath, it's executing serious business logic.
Conversational primary surface
The agent is the app. UI menus exist as a fallback, but most operators run the business through chat.
Production ERP underneath
CRM, jobs, accounting (double-entry, GL), inventory (multi-warehouse, FIFO), billing, HR, analytics — all built, not stubbed.
Industry presets
One-click configurations for landscaping, roofing, HVAC, consulting. Agents come pre-tuned to the vertical.
Self-modifying agents
Single-tenant deployment means the agent can actually edit the schema, run migrations, and change UI for that customer — without affecting any other tenant.
Construction vertical
Site inspections, safety incidents, ABC Supply integration, supplements/insurance claims handling.
Not a chatbot. An agent with real privileges, real tools, and real responsibility for running operations.
Function-calling against the whole stack
127 typed operations. Lead create, invoice generate, inventory adjust, schedule update, payroll run.
Self-modifying behavior
Per-customer agents can read/write schema, run migrations, ship UI changes — single-tenant means no blast radius.
Skill-based personality
Warm, consultative tone. Reads more like talking to a smart ops manager than typing into a CLI.
Multi-LLM routing
Gemini 2.5 Flash for everyday work, Claude for harder reasoning. Smart provider selection per task.
Actually does the work
Other "AI CRMs" answer questions. DinoClaw books the appointment, generates the invoice, updates the schedule, runs the accounting entry.
Single-tenant unlocks self-modification
Customer wants a custom field on every invoice? The agent ships it. No other platform can do that without a release cycle.
Real accounting
Double-entry GL with revenue recognition baked in — not a "lite" accounting feature, the real thing.
Frontend
Backend
AI
Infra
Scope
127 agent operations · 8,250 TS files (gateway) · 107K-line 3D command center · Per-customer isolated deployments
Metalcrafters
Turn job-site meetings into a production schedule that updates itself.
Seegars CRM
Replaced a 1990s desktop CRM with an AI-native cloud platform.
Cougar Companies
A 57-year-old contractor's digital nervous system — bidding portal, lead AI, content engine.
Want this kind of system for your business?
Tell me what you're running and what's slow. I'll send back a custom proposal with what I'd actually build, including price.