Watch it work
You stay in the loop: reads stay visible, and nothing changes in ConnectWise until you approve it.
Monday, August 31
Or open a Recent chat, or try Projects, Artifacts, and Schedules in the sidebar.
Every data point in this demo is fictional, for illustration only.
Paid package
Paid Pulsatrix includes the full stack, ConnectWise PSA, RMM, CPQ, ScreenConnect, SentinelOne, and Axcient x360, on the same local MCP host. Writes and higher-risk actions follow the approval policy you control. There's no à-la-carte plugin menu: paid means everything.
Service and business operations. Tickets, time, agreements, invoices, projects, sales, procurement, and team scheduling. 100 tools, one plugin.
Device and fleet operations. Resolve companies and devices, rank patch and CVE risk, audit policies and maintenance, then run a curated automation under approval.
Quote intelligence. Open and stuck quote portfolios, win-rate analytics, MRR on quotes, then refine lines or record won, lost, or expired.
Remote endpoint operations. Find sessions, read live state, run curated presets or approved commands, deploy packages, create support sessions.
Security operations. Agents, EDR threats, MDR alerts, Deep Visibility hunt, policy and assets. Then approval-gated triage, isolation, and mitigation.
Backup posture and recovery evidence. Fleet health, device risk, job history, AutoVerify proof, vault reachability, and ticket candidates for failures.
A free version of ConnectWise PSA also exists, with limited (read-only) actions.
| What it can do | Read / Write |
|---|
Bring your own model
Pulsatrix runs any model with reliable tool calling. Hosted is the fastest start. Local keeps client data on hardware you own. Either way, writes follow the approval policy you control before gated changes reach ConnectWise.
Strongest general reasoning, set up in minutes. You pay per token and data transits the provider. The easiest way to start.
Hosted for low-risk reads, local for sensitive work. The same approval gates apply, whichever you pick.
Client data never leaves your office. Quality scales with the machine. Pick by your box:
| Model | Agentic | RAM | M3 Max | M3 Ultra | DGX Spark | RTX 6000 |
|---|---|---|---|---|---|---|
| GLM-5.2 | 82.8 | 512GB+ | — | — | — | — |
| MiniMax M3 | 80.5 | 256GB+ | — | — | — | — |
| DeepSeek V4 Flash | 79 | 128GB | 21.5 | 27 | 14 | 18 |
| Kimi K2.7 | 78.2 | 256GB+ | — | — | — | — |
| Mistral Medium 3.5 | 77.6 | 80–128GB | — | — | — | — |
| Qwen3.6 27B | 77 | 24GB | 20 | 32 | 12 | 46 |
| Gemma 4 31B | 76.9τ² | 24GB | 12 | 20 | 11 | 40 |
| DeepSeek V4 | 76.2 | 256GB+ | — | — | — | — |
| GLM-5.1 | 74.4 | 384GB+ | — | — | — | — |
| Kimi K2.6 | 74.2 | 192GB+ | — | — | — | — |
| Qwen3.6 35B-A3B | 73 | 24GB | 85 | — | 21 | 196 |
| MiniMax M2.5 | 72.8 | 128GB+ | — | — | — | — |
| Gemma 4 26B-A4B | 68.2τ² | 16GB | 56 | 62 | 85 | 139 |
| Devstral Small 1.1 | 53.6 | 16–24GB | — | — | — | — |
| Command A | — | 80GB+ | — | — | — | — |
| Command R7B | — | 8–12GB | — | — | — | — |
Agentic = SWE-bench Verified (blue) · Tau2 (amber τ², Gemma). Model names link to Hugging Face. Hover for type/license. RAM = practical quantized set. tok/s where measured; "—" = not measured. Frontier MoE needs multi-GPU / Ultra-class. Cohere Command* often without published Verified score.
Security & privacy
Pulsatrix isn't a cloud middleman. It runs on your own infrastructure, with your own LLM, and client data never has to leave your office. The keys, and the policy, stay with you.
Your ConnectWise API keys live in your own environment: an OS keychain or a local secrets file. Never on a vendor server. The Pulsatrix host reads them locally, only when it calls ConnectWise on your behalf, so they never cross the network to us or sit in any cloud database.
Your AI client already asks before it acts; Pulsatrix adds a second gate it can't skip, met with Touch ID, Windows Hello, or the iPhone app when you're away from your desk. You choose what needs a signature: destructive actions never exist as tools, and a revocable grant lets similar calls through without asking again. Each person enrols their own device and profile, and no MCP tool can rewrite or bypass the policy.
Many vendor products give you one shared API key, so you can't limit what a single technician can do. Pulsatrix limits it at the gateway instead: each person gets their own plugin allowlist, with specific actions turned off just for them, over their own MCP/HTTP instance where a typo can't silently widen access. Credentials stay scoped to that profile in the OS keychain.
Reads flow across your stack with no approval prompt. Writes and other higher-risk actions wait for your decision, made now or already covered by a bounded grant you issued and can revoke. MCP tools can't skip that gate.
Every tool call is logged with a timestamp, the action taken, and its result, so you can see exactly what happened. That log lives on your machine, in a local file you control, not on a Pulsatrix server.
Pulsatrix runs on your infrastructure with your own model, local or cloud. There's no vendor middleman in the path. That's true for every connector it drives: PSA, RMM, ScreenConnect, and the rest all run through the same local host, never a hosted Pulsatrix backend.
For the CEO
Use a local LLM if you have the hardware, a cloud model, or both. Pulsatrix connects it to ConnectWise: the model and the host stay your decision.
Pulsatrix is the ConnectWise connector, not a second product brain. Bring a local model if you have the hardware, a cloud model, or both.
Which model, which clients, and where it runs stay yours. Products that take those choices away look simple on day one and expensive on day 400.
Run it on your own machines, on cloud you own, or a mix. Approvals stay legible. Liability stays understandable.
Read-only desk tools on your ConnectWise board. No credit card. Upgrade later for writes and every plugin.
PULSATRIX