Keep approved source code inside a controlled route
Local code assistance measured against your repositories.
A local coding endpoint can reduce source-code disclosure and give a team a controlled model version. It still needs repository permissions, licence review and acceptance tests.
Use local coding AI where IP control and sustained developer usage matter. Retain hosted tools where their capability and terms better fit the project.
Workload before hardware
Describe the queue, the quality bar and the operating owner.
- 01 / Work unit
- Size from model, context, concurrent users and peak demand.
- 02 / Acceptance
- Agree quality, latency, citation or output checks before buying.
- 03 / Operations
- Plan updates, monitoring, support and a fallback route.
Define what the model may see
Repository, branch, dependency and secret boundaries need explicit controls. Never index `.env` files, credentials or unrelated customer repositories by accident.
A coding assistant is not a reason to weaken normal source-control permissions.
Demand shape
Peak demand can matter more than the daily average
- Work unit
- Tokens, frames, files or jobs.
- Duration
- How long one active job occupies capacity.
- Concurrency
- How many jobs overlap.
- Deadline
- Interactive response or queued completion.
Trace the work before sizing the capacity.
The queue, memory shape, data path and management route turn a broad workload name into a testable service.
Evaluate useful engineering tasks
Test explanation, navigation, refactoring, unit tests and bounded code generation on representative code. Score correctness, security and review effort.
Autocomplete speed and repository reasoning are different workloads and may need different models.
Acceptance bench
Quality and service conditions pass together
Concurrency can matter quickly
Several developers generating long code responses can consume context and KV cache. The expected team pattern belongs in the sizing test.
Independent GPUs can separate models or teams where that produces a cleaner service boundary.
Operating loop
The workload continues after the first demonstration
- Observe Demand, errors and resource state.
- Review Quality drift, access and incidents.
- Change Versioned model or runtime update.
- Retest Focused acceptance before wider use.
Human review remains part of delivery
Generated code needs normal review, tests, dependency scanning and secure-development controls.
The appliance provides capacity and access; it does not certify generated code as safe.
Questions answered
Straight answers to common questions
Can a local model understand our whole codebase?
Repository-aware tools can retrieve relevant files, but context and retrieval limits remain. Test the exact repository and tasks.
Will our source code leave the building?
A local-only route can keep approved code on the customer-controlled system, subject to connectors, telemetry and any hosted fallback policy.
Can it replace developer review?
No. Generated or modified code requires normal review, testing and security checks.
Continue the decision
Useful next steps
Put the claim to work
Turn this guidance into a testable requirement.
The brief asks about workload and operating conditions - not just budget.