Context Engine
Codistry can build an understanding of your whole project, so the agent finds the right files and grasps how your code fits together before it starts working: fewer wrong turns, more accurate results.
The Context Engine is enabled out of the box (adronite-agent.contextEngineEnabled defaults to true), so you get its benefits without any setup. For the best results, build an index once from the Index tab, see Building the index.
What you get
The engine works quietly in the background and gives you four things:
Code understanding
The agent learns how your code connects (what calls what, what depends on what) so it can trace impact and navigate confidently.
Find by concept
Locate where a feature lives by describing it in plain language, even when you don't know the file names.
Automatic context
The agent orients to the most relevant files at the start of each task, and can pull more on demand as it works.
File outlines
A quick outline of a large file (its classes and methods) so the agent finds its way without reading the whole thing.
Understanding your code
Codistry builds a map of how your project's pieces fit together. With it, the agent can answer questions that span many files instead of guessing from one, for example:
- What would this change affect before I make it?
- Where is this function, class, or type actually used?
- How are these two parts of the code connected?
- What does this unfamiliar codebase look like at a high level?
- Which modules and directories depend on which?
It works across many languages, with the most depth in TypeScript and JavaScript.
From the Index tab you can hit Visualize to see a graph of your codebase. It's there to give you a feel for what the engine has mapped, not a polished diagram built for careful human review, so treat it as an illustration rather than a precise reference.
Finding things by concept
Describe what you're looking for in plain language, “where do we handle password resets?”, and Codistry points the agent to the files most likely involved. It's ideal when you know what a feature does but not where it lives.
Finding things by concept works best once you've built an index for the project, see Building the index.
Automatic context
At the start of a task, Codistry can hand the agent a short list of the files most likely to matter, so it begins in the right place instead of searching blindly, and it can refresh that list on demand while it works. These are pointers to relevant files; the agent still reads the actual code itself.
It can also surface files that tend to change together with the ones you're working on, so related code doesn't get missed. You can tune this behavior with contextPackAutoInject and contextPackCoChange.
Building & maintaining the index
Finding by concept and automatic context get noticeably better once you've built an index of your project. Build it from the Index tab in Codistry's settings, it runs in the background, so you can keep working while it finishes.
Stays up to date
Once built, the index refreshes automatically as you commit, check out, or pull, keeping pace with your code without you thinking about it. Prefer to be asked first? Turn off contextIndexAutoUpdate and Codistry will prompt you after a commit instead.
Privacy
With the default settings, building and using the index happens locally. No code is uploaded to make it work. Telemetry never includes your source code, file contents, or prompts.
Settings
Every option below lives under the Context Engine section of the Settings Reference.
| Setting | Default | What it does |
|---|---|---|
contextEngineEnabled | true | Turn the Context Engine on or off. |
conceptSearchEmbedder | local | Build the index locally or with a cloud provider. |
contextPackAutoInject | true | Give the agent a head start of relevant files each task. |
contextPackBudget | 8000 | How much relevant-file context to gather per request. |
contextPackCoChange | true | Also surface files that usually change together. |
contextIndexAutoUpdate | true | Keep the index fresh as your code changes. |
contextIndexNotifyOnCommit | true | Prompt to refresh after a commit when auto-update is off. |
contextIndexModel | default model | Choose which model builds the index (e.g. pin a cheaper one). |
fileBriefEnabled | false | Let the agent get a quick outline of a large file. |
The Context Engine is under active development. The defaults are safe to leave on, but its behavior may change between releases. Treat its suggestions as a fast way to orient. The agent always confirms against your actual code before acting.
- Settings Reference: Context Engine
- Expertise, the who-knows-what profile mined alongside the index
- MCP Extensions, add more tools the agent can use
- Best Practices, getting precise context into prompts