Waydev Brain · Powered by Waydev 15B
Waydev Brain helps engineering leaders ask questions across delivery performance, AI adoption and engineering investment. With a privately deployed Waydev 15B configuration, inference runs in your controlled environment rather than through an external model API. Deployment architecture and data flows are reviewed with your security team.
15B SDLC model
Waydev 15B deployment: 9 in 10 answers right in Waydev internal tests · 53% fewer incorrect answers than the base model · One 16 GB GPU minimum · No per-token model API fees. Review the benchmark methodology.
Waydev-reported evaluation: approximately 9 in 10 answers correct, with 53% fewer incorrect answers than the same base open-source model before domain training. The release benchmark has 104 frozen cases, repeated three times per case; numeric answers are checked against reference queries. These are Waydev’s internal test results, not an independent third-party benchmark.
How Waydev tests each release · Hardware requirements · Request an enterprise demo
Engineering leaders should be able to ask “Why did cycle time go up last sprint?” and get a trustworthy answer from their own engineering data. Sending sensitive repository metadata, tickets, developer activity or internal delivery metrics to an outside AI API is not acceptable for every organization.
| Team | Problem today | With Waydev Brain |
|---|---|---|
| Engineering leaders | Need an analyst or custom report | Ask questions in natural language and receive data-backed metrics, charts or widgets. |
| Security and IT | External AI services may be blocked | Run Waydev 15B locally on your GPU without external model-provider calls in the air-gapped configuration. |
| Works Councils and HR | Individual performance information needs protection | Works Council mode supports team-level responses and anonymization controls. |
| Finance | Variable third-party model usage fees | Local model inference with infrastructure costs instead of per-token model API fees. |
| Platform teams | Generic AI guesses metric and schema definitions | Waydev-domain training covers metric definitions, data models and supported tools. |
Engineering data foundations: see the Waydev SDLC documentation for the engineering data context and the Agent guide for asking questions of it. Dashboard access can be public within a workspace or private with explicit per-user view and edit permissions, as described in the dashboard access documentation.
Related capabilities: Signals, Goals, and AI coding ROI research.
Product direction · Exploring
Every pull request, issue and deployment tells part of the story. The reasoning behind decisions, constraints and trade-offs often lives in conversations or AI coding sessions that are harder to find later. Our vision for Waydev Brain is to connect the what changed with the why it changed across engineering teams and tools.
Example: If a billing change causes a regression, imagine asking Brain which prior changes affected the calculation, what related issues and tests revealed, and how the change affected delivery metrics—all with links to the underlying evidence.
Forward-looking concept: This section describes a product direction, not a currently available feature. Waydev Brain is not represented here as recording AI agent sessions, syncing prompts into Git or providing repository-backed persistent memory today.
Waydev 15B is specialized using questions from engineering analytics, Waydev metric definitions, the Waydev data model and privacy requirements. According to Waydev’s technical training process, each model iteration uses approximately 1,500 checked conversations. Training examples with numerical results are validated against reference queries; examples that fail the checks are excluded.
Waydev-reported results: approximately 90% correct answers and 53% fewer wrong answers than the unmodified base model on its internal engineering-question benchmark.
Waydev’s release gate uses 104 frozen test cases run three times per case. Each release must outperform the current version without any tested skill area dropping by more than 10 points. Numerical outputs are verified against reference queries rather than by visual judgment. Waydev 15B is currently described as version 3.
These figures describe Waydev’s internal evaluation under its stated test conditions. They are not a guarantee of accuracy for every customer dataset, language or workflow.
Architecture and data-egress requirements should be validated against the specific customer’s network and integration setup. See Waydev Security.
| Requirement | Waydev Brain support |
|---|---|
| Third-country data transfer control | Local inference keeps model requests on customer-controlled infrastructure when configured for air-gapped operation. |
| No individual rankings | Works Council mode is intended to keep responses at team level and refuse developer ranking requests. |
| Data minimization | Anonymization can hide colleague names so reporting emphasizes team patterns. |
| Transparency and accountability | Model, skill and user are recorded for each AI request. |
Works Council mode is configured at account level and may be documented in an employee-representation agreement. These features support governance but do not replace a GDPR, employment-law or EU AI Act assessment.
For enterprise deployments, Waydev can tailor a private Waydev 15B model to your engineering terminology, metric definitions, technology stack, processes, documentation and privacy rules. A custom model is evaluated against Waydev’s benchmark plus customer-specific tests.
| Deployment | GPU memory | Context window | Parallel requests |
|---|---|---|---|
| Minimum | 16 GB, NVIDIA L4 class | 64,000 tokens | 4 |
| Recommended | 48 GB, NVIDIA L40S class | 128,000 tokens | 6 |
Linux, CUDA 12+ and 30 GB available disk space. Internet access is not required for local inference. Actual throughput and context availability depend on the configuration and workload.
Waydev 15B is specialized for engineering analytics questions, metric definitions and software delivery data. It is not intended to replace a general-purpose model for every task. Depending on your organization’s policies and configuration, other model providers may also be available in Settings.
No. In an on-premises deployment, customer prompts and engineering data are not sent back to Waydev for model training.
Not at a practical production inference speed. A supported NVIDIA GPU with 16 GB VRAM is the stated minimum.
English is standard. Other languages can be considered as part of a custom build.
Enterprise pricing is private and depends on deployment, scale and customization. There are no third-party per-token API fees when running local Waydev 15B, but infrastructure, support and licensing costs still apply.
See what your engineering leaders can ask, how answers are grounded and how the model runs in your environment.
