Internal AI assistant agency
AI assistants that actually answer your team's questions
Morsof builds internal AI assistants for SMEs. Based in Casablanca, working with clients internationally. RAG over your company docs, AI helpers in Slack and Teams, customer-data Q&A, drafting and classification, voice and text interfaces. Self-hosted, with full data residency.
Who this is for
Three signals that an internal AI assistant will pay back fast.
Leaders watching the team re-ask the same questions every week
Onboarding, policy, customer history, product details, past incidents. The answers exist somewhere in Slack, Notion, the CRM or shared drives. Nobody finds them. You need an assistant that does.
Ops teams stuck on lookup, drafting and classification
Agents look up customer info across systems, draft the same emails dozens of times, and classify tickets manually. You need an assistant that does the rote work and leaves the judgment to humans.
Tech leaders who need data-resident, self-hosted AI
You can't put company documents or customer data into a public chatbot. You want AI that runs in your environment, uses your own API keys, and leaves a clean audit trail.
What we build
Concrete deliverables, not slideware.
- Internal AI assistants with RAG over your company docs (Notion, Google Drive, SharePoint, Confluence, custom)
- Slack, Microsoft Teams, Telegram and web-based AI helpers tied to your team's existing surfaces
- Customer-data Q&A over the CRM, ERP and ticket history with strict access control
- AI drafting and classification: replies, summaries, tickets, briefs, against your tone and templates
- Voice interfaces for hands-free ops (calls, warehouse, field work) when text is impractical
- Self-hosted vector stores, audit trails and access control built in from day one
How we work
A four-step engagement designed to ship in weeks, not quarters.
- 1
AI assistant review
30-min callWe map the team's daily questions, where the answers live today, the surfaces (Slack, Teams, web, voice) and the data residency constraints. You leave with a 1-page recommendation, even if you don't engage us.
- 2
Assistant blueprint
1 weekWe deliver a written blueprint: data sources and access model, RAG architecture, model choice and fallbacks, surfaces, evaluation plan, hosting plan, scope, timeline and fixed quote. You decide whether to proceed before any code is written.
- 3
Build & evaluate
2 to 6 weeksWe ingest your documents, build the assistant, wire it into Slack or Teams or web, and evaluate against a real question set with your team. We deploy to your environment, instrument logging, and hand over documentation.
- 4
Operate & extend
OngoingWe monitor quality, refresh the document corpus, swap to new models as they ship, respond to incidents, and ship 1 to 3 enhancements per sprint. Cancel any time.
What we work with
We bring the platform expertise. You keep ownership of the system.
Models
- Anthropic Claude
- OpenAI
- Google Gemini
- Llama, Mistral, Qwen (open source)
- Whisper (voice)
RAG & data
- Postgres pgvector
- Supabase Vector
- Pinecone
- Qdrant
- OpenSearch
Surfaces
- Slack
- Microsoft Teams
- Telegram
- Web apps & chat widgets
- Voice (Twilio, custom)
Hosting & ops
- Self-hosted on hardened VPS
- Docker
- Audit trails
- Access control
- Monitoring & evals
What we've shipped
A few examples from the broader portfolio.
AI-Powered ATS
Internal AI assistant for hiring teams. Pulls candidate data from job boards and the ATS, scores against the role brief, and routes shortlists to managers via Slack and email. Cut weekly hours on resume triage.
Internal docs Q&A
AI assistant over a SaaS company's product docs and runbooks. Embedded in Slack, answers support and onboarding questions in seconds, with citations back to the source documents.
Frequently asked questions
What buyers usually ask before engaging.
How long does an internal AI assistant project take?
Most engagements ship a working assistant in 2 to 4 weeks and reach steady state in 4 to 8 weeks. The first useful version (RAG over a defined doc corpus, in Slack or web) lands fast. Voice interfaces, multi-system search and complex tool use sit at the longer end.
OpenAI vs Claude vs Gemini vs open source: how do you pick?
We pick based on quality, cost, latency, data residency and tool use needs. Claude tends to lead on reasoning and long context. OpenAI is strong on tool use and ecosystem. Gemini wins on cost at scale. Open source (Llama, Mistral, Qwen) wins when full data residency or air-gapped deployment is mandatory. We design so the underlying model is swappable.
How do you handle our data and privacy?
Your documents and queries stay inside your environment. We default to self-hosted vector stores (Postgres pgvector, Supabase Vector) and route LLM calls through your own API keys. We do not train models on your data. Audit trails and access control are built in from day one.
How do you price internal AI assistant projects?
Fixed-price for the build phase based on the blueprint scope, then a monthly retainer for hosting, monitoring and continuous extension. Most SME engagements land between EUR 5,000 and EUR 25,000 for the build and EUR 800 to EUR 2,500 per month ongoing. LLM API costs are passed through at provider rates.
What does ongoing maintenance look like?
We monitor assistant quality, evaluate new models as they ship, refresh the document corpus, respond to incidents, run weekly backups, and ship 1 to 3 enhancements per sprint. You get a shared Slack channel and a monthly written report.
Do you sign NDAs and handle our data carefully?
Yes. We sign NDAs on request, keep customer data inside your environment whenever possible, encrypt secrets at rest and in transit, and document every data flow. We do not train AI models on your data.
Where are you based, and do you work with international clients?
We are based in Casablanca, Morocco, and work with SMEs internationally. We operate in English, French and Arabic, and deliver remotely with regular video syncs.
How do I get started?
Book a 30-min AI assistant review. You'll leave with a 1-page recommendation tailored to your stack, even if you don't engage us. If we're a fit, the blueprint phase begins the following week.
Stop watching your team re-search the same answers.
Let's design the AI assistant that turns your company knowledge into instant answers.