Then
Human-limited execution
Velocity capped by how much one team could type, review, and context-switch.
Human-centric AI systems
Software no longer needs to be typed by humans. But human software engineering has never been more important.
AI accelerates how fast code is created. But humans still need to supply everything else: starting with intent, taste, judgement, and oversight. The missing piece between humans and AI has been a code agent management platform, where alignment, execution, and intelligence are wired as one system. Level Up Labs is building that stack.
The shift
Coding has gotten faster. Code agent management hasn’t.
Level Up Labs treats agentic work as a logical flow: who decides what, why it changes, and what gets shipped. End to end.
Now
AI-limited oversight
Models generate code and drafts at speed, but intent fragments across chats, tickets, and repos.
Next
Human-AI coordination layer
The bottleneck has moved from fingers to alignment. Systems must carry goals, rules, and tests together.
The stack
One verticalized system, not a horizontal maze of syntax
The Level Up Labs agentic coding stack layers integrate vertically. Our foundation, Pinion, grounds the design. Our products are how customers benefit from the outputs of this stack today.
Coordination plane
Plan and sequence with agents without losing the thread
Verified GitHub linkage, shadow workspaces, and a deterministic operational loop—list → frontier → go → release → retro—shared by humans and MCP agents.
Execution plane
Backlog goes to zero by being executed, not just discussed
Jira backlog items become Git-backed Ansible for AAP—Compose deployment, compose-smoke readiness, retrieval proofs, MCP tools wired to the health surface.
Narrative plane
Agents communicate truth, with receipts
Diff→claim mapping, semantic change detection, and trust scoring make agent-generated software legible and verifiable in 30 seconds.
Runtime plane
Agents agree on reality at runtime
A declarative runtime contract for AI-built systems: services, dependencies, health, and validation gates that compile to Podman/Quadlet deterministically.
Governance plane
Don't merge what your team doesn't understand
Merge governance, recoverability scoring, and an ownership audit layer—so high-velocity, AI-assisted codebases stay legible as they evolve.
Foundation
The hub: Pinion
The category story we are betting on is not “better coding models”, it is better coordination between humans and agents using the coding models we already have. Pinion is our Planning, Coordination, and Execution Engine: the system that lets teams plan, sequence, and ship with agents without losing the thread. Pinionize is the hosted productization of Pinion; the rest of the portfolio plugs into the same coordination model.
Across our projects we are seeing velocity and code quality that we believe is unusual—and potentially category-defining when paired with disciplined product scope.
For a concise framing of why this matters, see Code Agent Management Theory.
Products
Five products, one coordinated system
Each product covers a distinct plane in the agentic delivery loop—plan, execute, narrate, run, and govern.
Coordination plane
Pinionize®Closed Beta
Hosted control plane for Pinion: link a GitHub repo, plan and execute sprints with humans and agents through a verified shadow workspace, ship via PRs.
Maintainer-verified repos, planner/executor LLM tiers (plan-sprint, go), MCP and web parity over the same HTTP API, deterministic operational loop: list → frontier → go → release → retro.
Execution plane
BacklogZero®General Availability
Coordinates Jira backlog into Git-backed Ansible for AAP—Compose deployment, compose-smoke readiness, retrieval proofs, MCP tools aligned with health APIs.
product.md defines Jira→playbook→GitHub→AAP execution; Compose packaging and operator flows are documented under backlogzero; compose-smoke.yml waits for GET /api/ready then runs smoke.sh; README documents retrieval_proof pytest matching CI; agents-mcp.md maps MCP tools to GET /health, GET /ready, and GET /service_status.
Narrative plane
Factimonious®Beta
A standard for evidence-backed agent narrative—every claim links back to commit, file, and exact line range so agentic software is legible and trustable in 30 seconds.
Diff→claim mapping, semantic change layer, deterministic + LLM hybrid (LLMs describe truth, they don't create it), trust scoring, and `facter-bot explain <repo>` as the developer surface.
Runtime plane
PodbayOpen Core
Runtime contract layer for AI-built systems: declarative services, dependency graph, validation gates, and deterministic deployment to Podman/Quadlet.
podbay.yaml defines services, dependencies, health, and requirements; CLI flow init → validate → deploy → explain → diff; contract > configuration, validation > execution, agents agree on reality.
Governance plane
GhostShipOpen Core
Don't ship AI-assisted code your team doesn't understand. Merge governance, recoverability scoring, and a human understanding layer for AI-era development.
PR Gate enforces understanding as a merge requirement; Recoverability score (0–100) tracks concept drift, architecture drift, workflow expansion, risk pressure; ownership audits validate who actually understands each system.
Foundations
What sits under the product lines
Research and infrastructure that made the stack possible—anchored on Pinion.
Pinion (Internal Project)
Planning, Coordination, and Execution Engine—the hub of our stack. Pinionize is the hosted productization of Pinion; the rest of the portfolio plugs into the same coordination model.
Read the story →Crumbify (Internal Project)
Breadcrumb-based signal for the noisy web—human presence and inspectable trails over content volume. An early proof-of-thought example for Pinion.
Read the story →QandAgent (Internal Project & General Availability)
Own your intelligence layer: secure, local Q&A over internal docs—built by Level Up's Agentic AI practice, with no proprietary cloud LLM lock-in.
Read the story →H-Frame (Open Source)
Repository isolation topology for AI-assisted software delivery: disposable agent workspaces, a deterministic in/out membrane, and host-controlled export policy (Apache 2.0).
Read the story →Proof
Concrete shipped code, not slideware marketing promises
We ship in public where we can. Demos and repos are the receipts.
Velocity
18 → 1
One internal project had eighteen sprint planning-execution-retro iterations compressed into a single calendar week on the internal agentic build. Same scope, with far fewer handoffs, and measurable commit velocity x quality increases.
Method: Co-planned sprints, Pinion-driven coding and agent orchestration, tests and CI as gates.
Codebase
Podbay: Its first 6 days of life, 21 Pinion sprints completed. Sprint history
Release and stability
- Version: v2026.5.1
- Stability: public preview
- Contract status: evolving
- Receipt format: versioned
- Production claim: suitable for narrow Podman stacks, not a Kubernetes replacement.
v2026.5.1 is the current public May 2026 preview release (v2026.5.0 was the first). It is usable, but the podbay.yaml contract is not yet 1.0-stable.
Follow along
- GitHub orgRepos, issues, and release notes.
- Product indexWhere each layer shows up today.
- Request info & updatesProduct news, demos, and stack updates from the team.
Stay in the loop
Request more info and product updates
Tell us who you are and what you're exploring. We'll share news on the stack, betas, and demos—no spam, no mailing-list theater.
Follow the build. Stress-test the stack.
Star the repos you care about, read the foundations, or reach out if you want a guided pass through the coordination graph.