What's Codpal

Codpal is a tool that allows teams and companies to utilize company knowledge across LLMs through any platform/app of their choosing.

The knowledge layer is known as Cortex. It stores context across every issue, code-change and interactions to be referenced by future tasks. It handles relevance, recency and conflicts. Cortex allows companies to own, sell, rent and trade knowledge however they please thus turning knowledge into a valuable asset the company owns aside from the product/services it offers. Some benefits of having a tangible knowledge base include:

• Reduction on reliance on single individuals when unavailable.
• Company culture, rules and structure is maintained when ownership of company is transferred.
• Companies can rent/sell their knowledge to other companies/entities on demand. Company accumulated practices, patterns and findings are valuable assets that are lost and unaccounted for.
• New companies who want to imitate the practices, knowledge and experience of other companies are able to do so without spending years in the market effectively leveling the playing field against their competitors.

Cortex solves a critical issue currently experienced by many engineering teams across the world:

• Knowledge split across different LLMs on local machines. Many people use different LLMs on different machines. Conversations and tasks done on one LLM on one machine are limited to the LLM on that machine only.
• Knowledge is fragmented across multiple apps/platforms. Engineering teams use apps such as Linear, Jira, Clickup for project management, Notion and other note apps for documentation, Slack, Telegram, Discord, and other messaging channels for simple chats and conversations, Github and Gitlab for hosting their codebases, Figma for their designs and so on. The fragmentation of knowledge means that one would have to hop around multiple apps to get a complete idea of a specific concept, always making onboarding expensive and the resignation of employees extremely painful.
• When employees are unavailable the knowledge they have accumulated is lost or unavailable. Companies and teams now either have to relearn how the specific person implemented those things or reach out to that specific person to handle the issues. This becomes extremely important during sudden catastrophic failures and unexpected events where said person is suddenly unavailable and can't be reached.
• Reliance on single LLMs and tools. Overtime as teams use specific products primarily, they are passively being opted out of new tools and products that provide better value. This can be seen in current LLMs and their memory model, apps and platforms with their context, and even IDEs with their "flows". With current "security" risks and local LLMs becoming more and more competent, the edge commercial LLMs have is fading out and companies interest in owning their own "AI" is suddenly becoming more realistic.

Cortex solves these problems by providing a central memory layer that can be referenced by any LLM with simple and intuitive methods. The "Knowledge" stored by Cortex is owned by the company and can easily be referenced by anyone with access to it at any time. Cortex is able to ingest conversations, specifications, documentations, code practices and patterns that companies have built up over time and reference it on demand. It also takes in the roles and responsibilities of every team member for possible interaction purposes.

LLMs and agents can easily interact and pull from Cortex before they handle/take on any task. This allows companies to use and experiment with multiple LLMs without the concern for context loss. Companies can even start utilizing local LLMs for various tasks thus reducing their dependence on commercial LLMs and reducing their security concerns regarding sharing company info.

Codpal is to Cortex as a human hand is to a brain. Codpal allows Cortex to connect with multiple platforms including- Linear, Github, Slack, Notion, etc. Codpal takes in the information on these platforms such as issue specs from Linear, Codebase from Github(this can also include commit messages, PRs, code owners), Slack conversations on public channels, Notion docs and so on. These informations are then ingested into Cortex and mapped as entities (person, tool, repo, concept), relationships (owner, author, prefer, dependency) and memory (permanence, durability, importance, access count, confidence).