AI Division
Answers from your own documents, kept in house.
Your policies, manuals, contracts, old proposals and product data already hold most of the answers your staff look for. We build private assistants that search them, answer with a link to the source, and only show people what they are allowed to see.
In short
What does Private Knowledge Assistants from Code Colonies cover?
Answers from your own documents and data, kept inside your company. Deliverables include connected knowledge sources; the assistant in the channels you choose; retrieval that respects permissions; test question set and quality report. The work is delivered under 2good.dev at USD 2 or INR 199 an hour, a 150 hour minimum, no advance and payment after you accept the work.
- How does the work run?
- Assess (1 to 2 weeks), pilot (2 to 3 weeks), production (rollout) and managed operations (monthly).
- Which tools and standards are used?
- Claude, OpenAI, Llama, Qwen, Mistral and LlamaIndex, chosen to fit your systems.
- Who owns the result?
- You do. Once you accept and pay, the documents, code and intellectual property from the work are yours.
- How do I start?
- Book a call or send a short brief. A specialist replies within one working day.
When clients call us
Situations we are asked to resolve
- New joiners take months to learn where things are kept.
- Sales and support keep asking the same few experts the same questions.
- Information is spread across SharePoint, Google Drive, Confluence and email.
- Company documents cannot go into a public AI tool.
Scope
What the work covers
- Knowledge mapping
- We find the sources that matter and how often each one changes.
- Retrieval design
- Chunking, embeddings and hybrid search set up for your kind of documents.
- Access control
- Existing permissions carry over, so people only see what they could already open.
- Cited answers
- Each answer links to the paragraph it came from.
- Channels
- Web, Microsoft Teams, Slack, WhatsApp or inside your own app.
- Private deployment
- Your cloud, a private tenant or your own servers, with open weight models if needed.
How it runs
A clear path, with gates
- Step 11 to 2 weeksAssessChoose the users, the sources and the questions that matter most.
- Step 22 to 3 weeksPilotA working assistant for one team, scored on real questions.
- Step 3RolloutProductionMore sources and teams, single sign on and monitoring.
- Step 4MonthlyManaged operationsKeep sources current, fill knowledge gaps and watch costs.
Deliverables
What you receive, and what changes
What you receive
- Connected knowledge sources
- The assistant in the channels you choose
- Retrieval that respects permissions
- Test question set and quality report
- Reports on usage and unanswered questions
What changes
- Quicker answers for staff and customers
- Shorter onboarding
- Fewer interruptions for senior staff
- Company knowledge kept private
- Typical stack
- ClaudeOpenAILlamaQwenMistralLlamaIndexpgvectorQdrantAzure AI FoundryAWS Bedrock
Start here
Talk to us about private knowledge assistants.
Tell us what you are trying to achieve. A specialist replies within one working day, and the first conversation is free.
Questions
Will our data be used to train public models?
No. We use enterprise model agreements that rule out training on your data, or run open weight models on your own infrastructure.
What if the assistant does not know the answer?
It says it does not know and points to the nearest sources. We log these gaps so your team can write the missing material.