Incident response
Ask "what's down?" and Percival checks every monitor, every service, and every backup job — then tells you what needs attention.
Not demos. Not generic chatbots with a logo slapped on. The AI agents we build are the same ones we use to run our own MSP — they open tickets, monitor infrastructure, onboard clients, and talk to customers on our behalf.
Every engagement starts with understanding where your team's time is going — then we build the smallest AI system that gets that time back.
Trained on your services, pricing, and FAQs. Qualifies visitors, captures leads, books assessments, and escalates to a human when needed — like the assistant in the corner of this page.
First-contact resolution for common questions. Integrates with your ticketing system. Knows when to escalate. Answers at 2am without waking anyone up.
Multi-step AI agents that take a trigger (email, webhook, form submission) and execute a workflow: fetch data, make decisions, update systems, send notifications.
Your documentation, your processes, your FAQs — embedded and searchable via a private AI assistant your team can actually ask questions to. Replaces "where's the runbook."
Percival — our own AI agent — runs on your infrastructure and talks to your RMM, ticketing system, monitoring stack, and backup server. Ask it what's broken. It knows.
Invoice processing, form parsing, contract review, report generation. AI reads the unstructured stuff and puts it in the right columns of the right spreadsheet.
Every project starts with a scoping call where we identify the one or two workflows that would most benefit from automation. We scope the smallest version that delivers value, build it, test it with your team, then iterate.
All AI runs on infrastructure you own or control — no proprietary lock-in, no opaque "AI platform" subscription. Models are interchangeable: Claude, Gemini, GPT-4, or a local open-source model if privacy requires it.
Tell us your use casetools connected to Percival, our own AI operations agent — running live in production.
Percival runs on a dedicated LXC container in our data center. It has 75+ tools connected to every system we operate — and we've been running it in production since early 2026.
Ask "what's down?" and Percival checks every monitor, every service, and every backup job — then tells you what needs attention.
Run /onboard and Percival creates accounts across 6 systems, provisions storage, and sends the welcome email — in under 3 minutes.
Connected to Prometheus, Grafana, and Alertmanager. Ask it why an alert fired and it reads the logs, metrics, and runbook — then suggests the fix.
What we build, what we don't, and where AI is actually worth the spend.
Percival is the AI operations agent we built and use to run our own MSP. He has 26 connected tools (Microsoft 365, Proxmox, Backup Server, Action1, Huntress, Zammad, Discord, Gmail, Drive, Wiki.js, and more) and answers questions like "what's the status of all client backups," "why did this alert fire," or "create a runbook for X" — pulling live data from real systems and writing the answer in your voice. Productized as Percival AI Concierge for clients, starting at $249/month. The same agent, scoped to your business and tools.
We design every AI deployment around that question, and the answer is usually no. Most of our agents route through a private LiteLLM proxy that lets us mix providers — Anthropic Claude for reasoning, Groq for speed, locally-hosted models on our own GPU for anything sensitive — without your data ever sitting in a third-party training corpus. For deployments where data is genuinely confidential (legal, healthcare, regulated finance) we run models entirely on-prem on Proxmox with GPU passthrough, no internet calls. Your data, your infrastructure.
Yes — that's one of the most common builds. We index your runbooks, FAQs, vendor list, password reset paths, onboarding docs, or any other knowledge base into a retrieval system, and connect a chatbot front-end (web widget, Slack, Teams, or Discord) that answers from those documents instead of guessing from a public model. Every answer cites the source so your team can verify. We use this internally — it's the system Percival uses to answer "how do we handle X" from our own runbook library.
Indicative ranges, scoped per project. A custom lead-capture chatbot on your website with a tuned prompt, lead routing, and a month of post-launch refinement typically starts around $2,500 fixed-price plus $99–$249/month to run. A document-aware internal knowledge base agent (the "ask our runbooks" use case) typically starts around $5,000 depending on document volume and integration complexity. A workflow automation agent (e.g., draft replies to inbound leads, escalate based on intent, file in CRM) typically starts around $7,500 and scales with the number of integrations and the depth of decision logic. We quote fixed-price up front after a free scoping call, and we'll tell you honestly when AI isn't the right tool for the job.
Building. The model providers (Anthropic, Google, OpenAI, Groq, plus open-weight models we host) are the engine — but the agents themselves, the tool integrations, the knowledge-retrieval layer, the prompts, the lead-capture logic, and the operational glue are all code we write and own. That's why we can scope and price fixed-price work, and why we can run sensitive deployments entirely on-prem when the situation demands it. We are not a SaaS reseller.
Yes — and that's most of the work. The repetitive, document-heavy, intent-recognition tasks AI is good at exist in every industry: dental practices triaging new-patient intake, contractors qualifying inbound bid requests, clinics handling appointment confirmations, retailers answering product-fit questions, restaurants taking reservation overflow. The pattern is the same: identify the most repetitive thing your team does every week, and build an agent that handles the boring 70% so people focus on the interesting 30%.