Trask Vela

Brain – methodology as code – 22 specialist AI roles, from business analyst to security engineer. Standards, templates and quality criteria, versioned and tested like software. Industry knowledge packs for banking, utilities, energy, automotive and healthcare.

Engine – orchestration – A delivery state machine that drives every story through the workflow on your own Jira, GitLab or GitHub. Background agents work one story at a time, with AI critic and judge loops inside each phase. Configurable per client and industry without code changes.

Modules – 19 capabilities – From bug triage to a complete delivery line. Each module works on its own, so you can start with one. Maturity is shown openly: Matured, Core or Preview.

Task management – Jira, YouTrack, GitHub, GitLab

Code repositories – GitHub, GitLab

Information intake – Confluence, Figma, MS Office, MS Teams

Testing – Vitest, Playwright, Robot Framework

Security tools – SonarQube, AISLE

AI models – Anthropic, OpenAI, Qwen – configurable per task

Agent harness – Claude Agent SDK; other harnesses on the roadmap

Runtime – TypeScript, Node.js, Neo4j, Qdrant, Fastify, OpenTelemetry; deployed with npm or Docker

Minimum to start – access to your code repository and task management system. Integrations to your own standard tools can be added.

Inside each phase, AI checks AI – An adversarial critic attacks every design and an independent judge decides. Code review runs as an audit-and-fix loop until approved. Tests must run and report pass or fail.

Between phases, people decide – G0 business approves analysis, architecture and mockups. G1 a reviewer approves each story’s design. G2 a human merges every merge request. G3 a human approves every release.

No runaway spend – Every loop has a hard retry budget; cost is tracked per phase and per ticket, with budget caps.;

Full visibility – Every story’s state is on your board. Every step is logged, every artifact versioned, with one auditable commit per task.

G0 Analysis – facts registry with cited sources, business requirements document, system architecture and ADRs, clickable mockups, open questions.

G1 Design – story specification and technical plan, UX screens, task breakdown, design-gate log.

G2 Code & tests – the code, code-review record, executed test reports, security assessment (OWASP).

G3 Release – release notes, deployment runbook, cost and effort telemetry, requirement-to-commit traceability.

Runs in your estate – on-premise or private cloud; you choose and approve the models and endpoints.

Every build checked – static analysis, OWASP checks per story, no unresolved high or critical finding without explicit risk acceptance.

Accountable agents – each AI role commits under its own git identity, with a full audit trail of who generated and who approved what.

Industry packs – regulatory requirements such as DORA and PSD2 applied as defaults.

Frameworks – built with NIST AI RMF, OWASP Top 10 for LLM applications and MITRE ATLAS in mind.