They lack awareness of your internal SDKs, architectures, infra, APIs, conventions, and domain logic. As a result, agents guess, hallucinate, or perform blind, expensive searches — producing code that “looks right” but breaks in production.
Codrel eliminates this gap entirely.
Today’s agents generate code without the deep, project-specific understanding that real engineers rely on.
This results in:
- Incorrect imports and signatures
- Misinterpreted architecture
- Generic or hallucinated APIs
- High token consumption
- Code that doesn’t compile or integrate
- Endless back-and-forth refinement
Example: You’re building on Nosana inside Kiro. Kiro has no idea how Nosana’s architecture works. It searches the internet, burns tokens, and still produces broken code.
The problem isn’t the model — it’s the absence of context. The critical knowledge is locked inside repos, docs, patterns, and the minds of your engineers.
AI writes production-grade code only when it understands your system like your senior engineer does.
That includes:
- Documentation
- Folder structures
- Configuration
- API definitions
- Domain logic
- Architecture flows
- Internal libraries
- Deployment + infra details
- Conventions and patterns
None of this is visible to an AI agent by default.
That is the gap Codrel fills.
Codrel is a context engine that ingests all your project knowledge, structures it into a RAG-ready format, and serves it to coding agents through MCP, APIs, and IDE extensions.
Codrel gives your AI agent complete understanding of your real tech stack.
When Codrel context is plugged into Kiro or VS Code, the agent immediately writes correct, architecture-aligned code — without hallucination or guesswork.
Codrel upgrades agents from “autocomplete with LLMs” to true stack-aware engineering assistants.
A unified flow:
Codrel ingests:
- Repositories
- Directories
- Files
- Docs and URLs
- Sitemaps
- API specifications
- Custom patterns
Codrel outputs:
.codrelknowledge state- RAG-ready embeddings
- Structured metadata
- Indexed, searchable chunks
- API-ready context for agents
This becomes the backbone for agentic coding in Kiro, VS Code, or any MCP-compatible environment.
Codrel consists of four major subsystems:
- Codrel CLI — ingestion and context generation
- Codrel Dashboard (Next.js) — visualization, management, and API
- Codrel MCP Server — standardized interface that delivers context to AI agents
- Codrel IDE Extensions — integration layer for Kiro and VS Code
Together, they form a complete context infrastructure for AI-driven software development.
Codrel’s ingestion and structuring engine.
It processes any knowledge source and generates a normalized .codrel knowledge base.
Example:
npx codrel ingest \
--token=<token> \
--repo <github-url> \
--dir <folder> \
--files <file1,file2> \
--sitemap <yml> \
--pattern <pattern>
Demo:
codrel ingest \
--token=Fssdsioadsngoadn124bsdsg \
--repo=https://github.com/vercel/next.js \
--dir=apps/dashboard \
--files=package.json,README.md \
--sitemap=https://docs.nosana.io/sitemap.yml \
--pattern="docs/**"Generated output:
.codrel/projects/<id>/
chunks.json
meta.json
state.json
This forms the authoritative knowledge layer for agents.
A Next.js platform for:
- Visualizing ingested context
- Inspecting individual chunks
- Managing multiple projects
- Controlling ingestion workflows
- Serving context via API
- Observing usage and system state
Acts as the standard interface layer between Codrel and AI agents.
Responsibilities:
- Retrieve the correct context at query time
- Provide structured chunks and metadata
- Enable retrieval, reasoning, and planning
- Ensure stack-aware coding
Supports:
- Kiro agents
- VS Code AI ecosystems
- Autonomous agent systems
- CLI-based AI developers
Brings Codrel directly into developer workflows.
Functions:
-
Authentication management
-
Running the MCP server
-
Injecting context into agent loops
-
Writing workspace instructions:
- Kiro →
.kiro/steering/codrel-tools.md - VS Code →
.github/copilot-instructions.md
- Kiro →
-
Managing added/removed Codrel collections
-
Enabling context-driven autocompletion + generations
This ensures agents operate with the same understanding as your engineering team.
| Layer | Purpose |
|---|---|
| CLI | Ingest + structure knowledge |
| Dashboard | Manage, inspect, and distribute context |
| MCP Server | Deliver structured context to AI agents |
| IDE Plugin | Inject context into Kiro/VS Code workflows |
| Agent | Generate accurate, stack-aware code |
Codrel becomes the foundational context layer for all AI-driven development.
AI agents fail because they lack system context.
Context determines whether the AI writes production-grade code or guesses blindly.
Codrel builds the full context pipeline: ingestion → structure → distribution.
CLI, RAG pipeline, dashboard, MCP server, IDE integrations — a complete stack.
Works for teams of any size, across any tech stack, with any AI agent.
- Eliminates hallucinations
- Improves code correctness
- Reduces debugging time
- Cuts token usage
- Speeds up onboarding
- Increases agent reliability
Codrel establishes the “context layer” for enterprise AI coding agents, similar to how Git became the version control layer.
Codrel gives AI coding agents deep, structured, stack-level understanding — enabling them to produce accurate, architecture-aligned code from the first attempt.