One workspace for understanding, reviewing and documenting code
CodePilot AI connects to your repositories, builds a searchable model of your software and puts that knowledge to work for every engineer on the team.
How CodePilot AI works
A retrieval-first pipeline keeps answers tied to your real code. Here is what happens between connecting a repository and reading an answer.
- 1
Connect repositories
Link a repository from your Git provider with read-only access. Select the branches and folders CodePilot AI should learn from and the ones it should ignore.
Read-only access, per-repository scope
- 2
Ingest and snapshot
A repository snapshot is stored in Amazon S3 and an event starts an AWS Lambda job that unpacks it and lists every file.
Amazon S3 and AWS Lambda
- 3
Parse and chunk
Source files are split along function, class and module boundaries so each chunk keeps its meaning. Tests, configuration and documentation are indexed with the code.
Structure-aware chunking
- 4
Embed and index
Chunks are converted to embeddings with a model on Amazon Bedrock and stored in PostgreSQL with pgvector, next to metadata such as path, language and symbols.
Amazon Bedrock and pgvector
- 5
Retrieve context
A question is embedded, matched against the index, and expanded with related files such as callers, tests and configuration.
Semantic search plus code relationships
- 6
Generate and cite
A Bedrock model drafts the answer, review or document from the retrieved context. Each statement carries a file and line reference you can open.
Grounded generation with citations
Six modules, one shared understanding
Each module uses the same index, so what CodePilot AI learns while answering a question also improves reviews, documentation and tests.
Ask
Natural-language questions across one repository or several, with cited answers.
Review
Automated first-pass review of pull requests and branches, ranked by severity.
Document
READMEs, API references, architecture notes and onboarding guides.
Test
Coverage gaps, edge cases and test skeletons for your framework.
Map
Service and module dependency maps, with the riskiest areas highlighted.
Measure
Insights on hotspots, review turnaround and knowledge concentration.
Trust is built into how it works
Engineering teams will only rely on AI they can check. These principles shape every part of the platform.
Every answer shows its sources
Statements link to the files and lines they came from. If the repository does not contain the answer, CodePilot AI says so rather than guessing.
Read-only by design
CodePilot AI reads code to build its index. It does not commit, push or modify your repositories.
Your code is not training data
The platform is designed around Amazon Bedrock, which does not use customer prompts or completions to train AWS foundation models.
Access follows your organisation
Workspaces, repositories and roles determine who can ask about which code, and actions are recorded in an audit trail.
Deployment options
Start with a managed proof-of-concept, and move to a dedicated environment as requirements grow.
Managed environment
Operated by Source Code Matters on AWS. The fastest way to evaluate CodePilot AI on a real repository.
- Quick to start
- Single-tenant data separation
- Support during the pilot
Dedicated environment
An isolated deployment for teams with stricter data residency or network requirements.
- Isolated network boundary
- Customer-controlled access
- Planned for later phases
Repository sources on the roadmap include GitHub, GitLab, Bitbucket and Azure Repos.
Common questions
Can't find what you need? Ask us directly and we will answer in the demo.
Which languages and frameworks are supported?
CodePilot AI is designed for repositories that mix languages. The first pilots focus on TypeScript, JavaScript, Python, Java and C#, with more added based on what pilot teams use.
Does CodePilot AI change our repositories?
No. Access is read-only. Suggested fixes are shown for you to review and apply yourself.
Where does our code go?
The planned architecture keeps repository snapshots, embeddings and generated documents inside AWS services in a single environment. The demo will walk through exactly what is stored and for how long.
Can it run in our own AWS account?
A dedicated deployment is on the roadmap for teams that need it. The proof-of-concept runs as a managed environment operated by Source Code Matters.
How are answers checked?
Answers are grounded in retrieved code and include citations. Pilot teams help us build evaluation sets from their own repositories so quality can be measured, not assumed.
See what CodePilot AI finds in your codebase
Book a walkthrough and we will run CodePilot AI against a repository so you can judge the answers, reviews and documentation for yourself.