$ waydev models –default waydev-sdlc-15b
Waydev SDLC 15B is our own AI model, fine-tuned on software delivery data. It plans its own tool calls, queries your Waydev data, and answers with real numbers, charts and the SQL behind them.
It’s now the default AI model in Waydev, on Waydev Cloud and on-prem. If you use Waydev today, you’re already using it. Prefer another provider? You can bring your own key at any time.
On-prem, it runs on your own GPU with no internet access. Prompts, data and answers never leave your network.*
.gif)
// model card
What’s inside. In one file.
name: waydev-sdlc-15b base: open-source, fine-tuned by Waydev params: 15B quant: 4-bit (~9 GB) context: 64K–128K tokens serving: OpenAI-compatible API egress: none # no outside AI vendors accuracy: 9/10 # hold-out benchmark vs_base: -53% wrong answers version: v3
// why we built it
Leaders want answers. Their data can’t leave the building.
General-purpose AI can answer “Why did cycle time go up last sprint?”, but only if your data leaves the company. Security teams, Works Councils and regulators often won’t allow that. And general models guess at your schema: they mix up lead time and cycle time, invent column names, and return confident, wrong numbers.
We built a model that knows engineering analytics, and runs where your data already lives.
// how it answers
Plans the query. Shows its work.
> Why did cycle time go up last sprint? · pick metric pr_cycle_time · query by team, last 2 sprints · break down coding / pickup / review / deploy ✓ answer + chart + editable SQL
Illustrative trace. Every answer reads your data through Waydev’s tools and permissions.
// what you can do
From one question to a full dashboard.
- ask
- Plain-English questions over your PRs, commits, issues, deployments, teams and AI usage. Answers as text or charts, with editable SQL. Docs →
- generate
- Describe what you want to track. Get a dashboard with the right metrics and widgets.
- explain / predict / improve
- Three AI actions on every widget: what it means, where it’s heading, and how to improve it.
- expand
- Click the sparkle icon to turn one chart into a full dashboard on the same topic. Docs →
// how we trained it
Verified data in. Benchmark-gated releases out.
- Real questions. Generated from live Waydev data across a dozen skill areas, from SQL and date ranges to Works Council mode.
- Checked answers. Numbers must match a reference SQL query. No invented columns, no leaked names. Failures are dropped, not hand-fixed.
- Step-level training. About 1,500 verified conversations, roughly 4,400 tool-call and answer steps per version.
- Release gate. 104 frozen cases, 3 runs each. Ships only if it beats the current model, with no category dropping more than 10 points.
// private by design*
Your prompts, data and answers stay inside.
On-prem, Waydev SDLC 15B runs on a GPU server inside your network, next to Waydev on-prem, served by an open-source inference server. It needs no internet access to answer.
your-network/
- works_council_mode
- Team-level answers only. Trained in, not a prompt, and tested on every release.
- permissions
- Reads data only through Waydev’s tools, with your existing access checks.
- audit_log
- Every AI request logged with model, skill and user.
Designed to support GDPR, Works Council and EU AI Act requirements. Prefer a hosted model? Switch to Claude, or bring your own key, at any moment.
// deploy on-prem*
One NVIDIA GPU. Fixed cost. No per-token fees.
| Minimum | Recommended | |
|---|---|---|
| GPU | 16 GB · L4 class | 48 GB · A40 / L40S |
| Context | 64K tokens | 128K tokens |
| Users | 4 | 6 |
Host: Linux · CUDA 12+ · 30 GB disk · internet not required. Updates ship as a new model file after passing the benchmark gate: swap the file, restart the server. Custom builds are available for your language, metrics, internal docs and works agreement.
Run Waydev SDLC 15B inside your network.
We’ll scope the deployment with you:
- Hardware sizing, from one 16 GB GPU up
- Installation next to Waydev on-prem, no internet required
- An optional custom build for how your teams work
Waydev Brain, the private assistant built on Waydev SDLC 15B.
* Waydev SDLC 15B is available on Waydev Cloud and on-prem. On Waydev Cloud, Waydev runs the model on its own infrastructure, with no outside AI vendor. On-prem, it runs on a GPU server inside your own network, with no internet access needed; GPU and deployment details apply to on-prem only.
