The best way to give context to AI is not to
@Neo, fix it!
Assigning a ticket, commenting on a pull request or asking for a code review are simple mechanics that team members perform several times a day, and to accomplish them, they need to understand the context around them.
Without the context and know-how of the team, actions are generic and irrelevant. The context discovery system is designed so that neo always has access to all the relevant information, across tools, even from different vendors.
Guided Context Discovery
Neo's context discovery is carefully guided to follow the same patterns an experienced team member would use. When evaluating a comment, Neo naturally traverses the relationships between tools and artifacts, understanding not just what is connected, but why it matters.
Deep Relationship Understanding
A comment exists within a thread, which is part of a pull request, authored by a specific person, reviewed by others, and linked to a project management ticket. Each connection provides crucial context that informs Neo's understanding and response.
Beyond Vector Databases and Fine-Tuning
While vector databases and fine-tuning are valuable tools, they have proven insufficient on their own. Vector similarity doesn't capture the structured relationships between artifacts, and fine-tuning alone cannot encode the dynamic, evolving context of a living project.
Practical Context Windows
Having the entire context available in every prompt isn't feasible due to token limits and cost constraints. Neo intelligently selects and prioritizes the most relevant context, ensuring that what matters most is always available when making decisions.
Not Just Agentic Tools
Unlike purely agentic approaches that rely on LLMs to decide which tools to call, Neo's context discovery is deterministic and structured. This ensures consistent, reliable context gathering without the unpredictability of agent-based tool selection.
When you write "@neo fix it", Neo already knows what "it" is. Neo understands the code being discussed, who wrote it, why it was written (from the ticket), what the broader context is (from linked documentation), and what similar issues have been encountered before (from the knowledge database).
This is how Neo can be so effective with such simple commands—the context is already there, discovered and understood through a carefully designed system that mirrors human expertise.