Contexts & Advanced RAG
Standard AI chatbots fall apart when dealing with highly specific technical issues because they lack the context of your business. A generic LLM cannot write a ticket to "Fix the authentication bug" because it doesn't know if you use Firebase, Auth0, or custom JWTs.
Plan AI solves this using Contexts and Advanced Retrieval-Augmented Generation (RAG).
What is a Context?
A Context in Plan AI is a defined boundary of knowledge that you explicitly attach to a meeting or a chat session.
You can create a Context for a specific microservice, a specific client, or an entire product. Inside that Context, you can upload:
- Architecture Documents (Markdown files explaining how the system works).
- API Schemas (Swagger/OpenAPI JSON files).
- Code Snippets (Specific components or functions).
When you generate a Jira ticket or ask the chat a question, Plan AI automatically fetches this Context and injects it into the prompt.
This ensures the LLM's output is grounded in your actual technical architecture, drastically reducing hallucinations.
Vector Search (RAG)
If your team has been using Plan AI for six months, you might have hundreds of hours of meeting transcripts.
When you ask the chat, "Why did we decide to use Qdrant instead of Pinecone?", the system doesn't just guess.
- It uses Qdrant (our Vector Database) to perform a semantic similarity search across all historical transcripts in that Context.
- It retrieves the exact 5-minute chunk of a meeting from three months ago where the engineering team debated the cost-efficiency of Qdrant vs Pinecone.
- It feeds that specific chunk to the LLM to formulate an exact, cited answer.
This gives your TPMs and developers perfect corporate memory.