Skip to content
donniedice / SOFTWARE

WORKFORCE BY DONNIEDICE

Your backlog.
A workforce
behind it.

Give your GitLab issues a team of specialized AI agents. Route the work, get changes back for review, and keep control of your infrastructure.

See how it works Find your models & API keys ↗

Infrastructure loadout. Eight roles. Existing-cluster and greenfield installation paths.

WORKFORCE ARCHITECTURE 8 ROLES
◇ Your GitLab backlog ISSUE →
DispatcherTriage · scope · route
COORDINATE
02Engineer
03QA
04Reviewer
05Security
06Platform
07Release
08Docs
Merge request + evidenceYour gates
YOUR INFRASTRUCTURE ILLUSTRATIVE FLOW / NOT LIVE ACTIVITY
08Specialist rolesSeparate identities & permissions
03Provider integrationsNVIDIA, Google & OpenRouter free routes
5 minQueue check intervalConfigurable; not a delivery SLA
BYOInfrastructure & keysYou choose providers and budgets

01 THE PACKAGE

The setup is the product.
The work stays yours.

A deployable foundation for people who already run repositories and infrastructure. The roles, routing, and operating guides come together in one loadout.

A delivery workflow,
not another chat window.

Move from a scoped GitLab issue to a task branch, tests, and a merge request. Keep assignments and completion evidence alongside the code.

ISSUEWORKREVIEW

Boundaries you control

Separate namespaces, persistent identities, and role-scoped credentials. Enforce authority with GitLab permissions and cluster policies.

Infrastructure as code

A reusable Terraform module, role configuration, staged deployment scripts, and GitLab CI checks you can inspect and adapt.

Start with or without a cluster

Use the existing-cluster path today. A provider-neutral VPS-to-K3S bootstrap lane is part of the launch target and remains gated on clean-machine validation.

Know the boundary. This is a self-hosted software package, not hosted compute, model credits, guaranteed unattended delivery, or a managed support service.

02 FROM ISSUE TO REVIEW

Set it up once.
Work from your backlog.

  1. 01

    Prepare your environment

    Bring the infrastructure and accounts you already run on. Guided setup prepares your environment for the workforce.

  2. 02

    Deploy & verify

    Validate the configuration, stage namespaces, provision private role credentials outside Terraform, then review and apply a saved plan. Run the health checks.

  3. 03

    Assign a scoped issue

    Write the expected outcome, acceptance criteria, and allowed scope. Assign an authorized role identity. Eligible work is checked on the configured schedule.

  4. 04

    Review the evidence

    Inspect the branch, merge request, and checks. Your configured review and release gates decide what can merge or deploy.

EXAMPLE WORKFLOWIllustration, not live telemetry
GITLAB / ISSUE

Fix the empty-state crash on the settings page

Include reproduction steps, expected behavior, and a regression test.

Scoped assignment
  1. 01Engineer
  2. 02QA
  3. 03Reviewer
  4. 04Release

A result you can inspect.

A proposed fix, reproducible checks, and a reviewable merge request.

Handoffs depend on issue metadata, role access, and the configured gates.

03 DIVISION OF LABOR

Eight roles.
Clear responsibilities.

Select a role to see what it contributes. The implementation identity and review identity have different jobs.

Coordinate / ROLE 01

A clear assignment for the right specialist.

Triages issues, identifies risk, and routes scoped work. GitLab stays the durable work queue.

EXPECTED OUTPUTScope, acceptance criteria, and a role assignment

THE DESIGN, MADE VISIBLE

Coverage. Continuity. Control.

Configuration diagrams, not model scores or performance benchmarks.

01 / ROLE COVERAGE

Separate the work from the checks.

  • Coordinate1
  • Build3
  • Check3
  • Deliver1

Engineer, Platform, and Docs implement. QA, Reviewer, and Security provide separate checks.

02 / MODEL OPTIONS

Start with a no-cost model baseline.

  1. 1
    NVIDIAKimi K3
    BYOK
  2. 2
    NVIDIAGLM 5.3
    BYOK
  3. 3
    GoogleGemini Flash
    BYOK
  4. 4
    OpenRouterFree catalogue
    BYOK

The supplied default names multiple providers. Quotas, availability, model suitability, and provider data terms must be checked for your account; compatible paid routes can be configured.

03 / DEPLOYMENT CONTROL

Checks before changes.

  1. 01Static policyDeployment stage
  2. 02ValidateDeployment stage
  3. 03PreflightDeployment stage
  4. 04Plan reviewOperator decision
  5. 05Saved-plan applyChecksum + age guard
  6. 06VerifyDeployment stage

04 MODELS & API KEYS

Bring your keys.
Use free AI well.

The shipped default runs on tested no-cost AI routes. The package price does not include hosting, and provider quotas or eligibility can change.

Bring your own model access.

Compare provider terms, quotas, and task fit before choosing a route. The public requirements overview covers what you need; detailed model selection and verification guidance belongs in the buyer package.

N

NVIDIA API Catalog

Dispatcher + worker primaries

Account-dependent trial routes

  • moonshotai/kimi-k3
  • z-ai/glm-5.3

These two default routes answered bounded requests on the reference account. NVIDIA trial endpoints log usage under the NVIDIA API Trial Terms; verify eligibility and review terms before sending repository content.

Join the NVIDIA Developer Program, choose a model, select Get API Key, and ping-test the exact model ID before configuring roles.

Open NVIDIA catalog
Setup instructions ↗Models / terms ↗
G

Google AI Studio

Dispatcher short-task fallback

Free tier where eligible

  • gemini-3.8-flash

Gemini Flash answered a bounded request on the reference account. Check the account-specific quota, session length, and data policies before using it for agent work.

Create or select a Google project in AI Studio, generate a key, and check your project limits and data-use terms.

Create a Gemini API key
Setup instructions ↗Models / terms ↗
OR

OpenRouter

Quota-capped free fallbacks

Free model options

  • z-ai/glm-5.2:free
  • qwen/qwen3.8-27b:free

GLM 5.2 Free and Qwen 3.8 27B Free extend the fallback chain. The :free variants enforce daily quota caps and may 429 during heavy use, so treat them as continuity routes, not primaries.

Create a key, filter the catalog for free models, and copy the complete model ID including :free.

Create an OpenRouter key
Setup instructions ↗Models / terms ↗

Model IDs above describe the shipped configuration, not guaranteed account entitlements. Every route was ping-tested against the current internal model comparison; catalog availability and quotas can change, and OpenRouter :free entries enforce daily caps. NVIDIA trial endpoints log usage under their trial terms. Provider documentation is the authority for quotas and data policies.

05 BEFORE YOU DEPLOY

Built for operators.
Owned by you.

Bring these foundations.

  • An existing Kubernetes cluster with persistent storage and suitable networking.
  • GitLab identities, project membership, and scoped permissions for the roles you enable.
  • Role-scoped model credentials and budgets for each configured route.
  • Terraform/OpenTofu, kubectl, Python, and someone responsible for operating the environment.

Be clear about the trade-off.

You gain control over the deployment and provider choices. You also operate the cluster, review generated changes, manage quotas, and maintain your access policies.

Agent output needs validation. Queue polling is scheduled; completion time depends on the task, model, and available capacity.

Read the requirements

06 STRAIGHT ANSWERS

A few things worth knowing.

What am I buying?

workforce: a self-hosted infrastructure and role-configuration package with eight role definitions, a GitLab dispatcher reconciler, packaged skills, deployment scripts, CI configuration, and operator guides. It runs as Hermes Agent gateways deployed by Terraform/OpenTofu. It is not a hosted AI subscription or a bundle of provider credits.

Can I run it with free API keys?

The default starts with no-cost routes that answered bounded checks on the reference account. Other candidates may have quotas or require different terms. Compatible paid models may also be configured; the buyer package includes verification guidance.

What if I do not have Kubernetes yet?

The product supports an existing-cluster lane and is being completed with a provider-neutral path from a customer-owned VPS to validated K3S. The greenfield lane will not be described as shipped until its clean-machine, reboot, storage, backup, and full workforce canary tests pass. Hosting and ongoing server operations remain customer-owned.

How does an agent know what to do?

You create a GitLab issue with the repository, scope, acceptance criteria, and checks, then assign it to an authorized workforce identity. The reconciler checks for eligible work on a schedule. Queue depth, worker health, project membership, approvals, and model availability determine when it can start.

Where do my keys and source code go?

You provision role-scoped credentials on your own infrastructure, outside Terraform state. Agents send prompts and relevant work context to the model providers you select and use your GitLab for durable work. Review provider data policies and grant only the access each role needs; self-hosting is not a promise that data never leaves the cluster.

Can it merge or deploy without review?

The supplied roles separate implementation, review, and release responsibilities. You must configure GitLab permissions, protected branches, and approval gates to enforce the authority you want. Prompts describe behavior; credentials and platform controls enforce it.

Are the diagrams performance benchmarks?

No. The diagrams show the package structure and an illustrative issue workflow. The five-minute interval is a scheduler setting, not a completion guarantee. We do not publish a task-success rate or a universal delivery-time claim.

Can I purchase workforce today?

If the Square checkout appears below, you can purchase at the total Square calculates. A private order link is issued after payment and activates when the package is published. If checkout is not visible, sales are not open yet; contact donniedice instead.

Joseph Gettings Jr, donniedice

BUILT BY AN OPERATOR

donniedice.

Joseph Gettings Jr designed this loadout from a role-separated Kubernetes and GitLab workforce. The package brings that operating experience to teams who want a self-hosted starting point; the detailed setup and verification material belongs in the buyer release.

Talk to the builder

WORKFORCE / GET STARTED

Put your backlog
in motion.

A self-hosted, eight-role setup for your GitLab work. Bring your infrastructure and model keys; the package includes the workforce configuration and operator guide.

Review what you need before buying

Checkout is currently unavailable. No payment is collected here.