Code Intelligence

Understand the change before you make it.

This is the project-reality layer of Strata19's AI-native IDE: durable local repository context, focused code relationships, bounded detector classes, work, checkpoints, and evidence. It does not ask a model to invent facts about your repository.

The working loop

Use repository intelligence at the moment a change matters.

This is the detailed workflow behind the homepage overview: understand, retain the decision and scope, then inspect evidence after the change.

01

Ask a bounded question

Inspect a repository, a file, or a symbol before you edit. The result should identify its analyzed scope and caveats.

02

Use the resulting context

Review callers, callees, references, and related files. Keep the useful result attached to the work rather than relying on chat memory.

03

Check again after the change

Compare what is reported after the change with a checkpoint. “Gone” means absent from the compared finding set, not proven repaired.

01 · Detectors

Four local checks, each with a specific boundary.

This is the first intelligence layer and it is available in Free Local today. Each detector creates a concrete review queue; none certifies a repository as correct, secure, or complete.

GC-01

Unreachable-file candidates

A conservative whole-file reachability pass from indexed entrypoints. A candidate is a starting point for inspection—not deletion authority.

How you use itUse it when planning cleanup: inspect the candidate, dynamic loading, generated code, and repository conventions before deciding whether to remove anything.

BoundaryIt reports at file level only. It does not prove an export is unused, and it intentionally stays silent when entrypoint or graph evidence is incomplete.

GC-02

Duplicate blocks

Exact token fragments and short function-body hash collisions, reported as a concrete duplication inventory.

How you use itUse the cluster to decide whether the copies should share an implementation—or whether their differences are intentional and should remain visible.

BoundaryA duplicate is a maintenance question, not an automatic refactor. It does not decide which copy is canonical or safe to change.

GC-08

Placeholders and stubs

TODO/FIXME markers, empty bodies, not-implemented throws, stub returns, and stray console logging.

How you use itUse it before declaring work complete: open the reported source location, classify the unfinished path, then create or update the work that addresses it.

BoundaryA marker is not necessarily a defect. Some logging and comments are deliberate; the detector supplies a concrete review queue, not a verdict.

GC-13

Configuration gaps

Missing environment examples, placeholder package names, missing required scripts, and malformed package metadata.

How you use itUse it when preparing a repository for another developer, CI, or a new environment: review the inventory and make the project’s expected setup explicit.

BoundaryIt checks a deliberately narrow local set of configuration signals. It does not validate deployed infrastructure or prove that every environment will start.

strata19 / verify implementation
local evidence

$ strata19.verify_implementation

GC-08 · placeholder candidate

GC-02 · duplicate block candidate

GC-13 · configuration gap

Read each finding with its source evidence and caveats.

What this does not say

  • The repository is secure.
  • Every issue has been found.
  • A candidate is safe to delete.

02 · Structure

Know the neighborhood before you touch a line.

Structure turns a request such as “where is this used?” into a deliberate review path. Free Local keeps focused context close to your repository; the connected layer is planned to extend the project-wide structural map.

Start at the relevant structure

Ask about a file or symbol before editing. The useful result is the nearby implementation and relationships—not a giant dump of search matches.

Follow references with intent

Use callers, callees, imports, exports, and related files to choose what needs review alongside the change.

Carry uncertainty forward

When a relationship cannot be resolved or the available context is incomplete, that caveat belongs in the result. Do not treat an absent edge as proof.

03 · Impact & risk

Turn context into a reviewable change plan.

The connected intelligence layer is planned to make a proposed change easier to scope before it lands. Its job is to show the relationships worth reviewing—not to promise that every downstream effect has been predicted.

Trace the change surface

Use the relationship map to identify the callers, consumers, and nearby contracts that may deserve review when one implementation moves.

Choose evidence to revisit

Turn the change surface into a practical review list: the relevant findings, checkpoints, and repository evidence to inspect again after the edit.

Keep the human decision visible

A risk signal helps prioritize review. It is not a deployment approval, a security finding, or an instruction to merge without judgment.

Planned with Connected · not part of today’s Free Local release

04 · Project layer

Keep the decision, work, and evidence after the chat ends.

The project layer is where one-off local context becomes durable project intelligence. Free Local keeps it with the repository; the connected workspace adds cloud-synced history and web-only views when it ships.

Retain the work

Work and project state persist locally so the reason for a change does not disappear with a single coding conversation.

Compare against checkpoints

Use a recorded finding set as a baseline, then inspect what is new, still reported, or gone in a later run. “Gone” means absent from the compared finding set—not proven repaired.

Shape routines by conversation

Describe recurring work to your assistant, then review the routine draft, validation, and dry run before publishing. Managed schedules, webhooks, and runners remain planned for Pro.

Connected and Pro are upcoming · the local plugin remains available without cloud sync

Upgrade boundary

Local intelligence first. Connected analysis later.

Free Local is the released local product, including conversational routine drafting. The upcoming connected plans add cloud-synced history, the web workspace and Engines; Pro is planned to add managed schedules, webhooks, and runners.

Available now

Free Local

Local repository intelligence, focused context, work, conversational routine drafting, checkpoints, bounded verification, and optional BYOK assistance.

Upcoming

Connected

Cloud-synced project history, the connected web app, Engines, and bundled plugin-agent inference.

Upcoming

Pro

Connected project intelligence with the planned managed-execution layer and a larger Strata Credit allowance.

Start with the repository you already have.

Install Free Local in a supported coding harness and ask a bounded question before your next change.