Spec v1.0Open Framework

SliceOpsThe decision-first framework for AI-first software engineering.

The decision-first framework for AI-first software engineering.

Multi-agent teams ship auditable software, not vibe code.

The problem

AI can write code,
can you trust its decisions?

Tools like Cursor, Devin, and Copilot solved generation speed. They created a new bottleneck: understanding, validating, and auditing the decisions agents make in the background.

When a dozen agents touch the same repository in parallel, who records the architecture? Where is the evidence a human approved it? How do you answer an EU AI Act or ISO 42001 auditor?

The framework

The audit plane

SliceOps is not a code editor or an agent. It is the open framework that sits above them — the layer where decisions are recorded, traced, and kept consistent across slices.

Decision integrity

Every slice produces a Decision Record. Architectural intent becomes an immutable, machine-readable trail — not knowledge trapped in a chat history someone has to reconstruct later.

Slice atomicity

One chat, one PR, one atomic slice. Scope is declared upfront and bounded to what an agent can carry end-to-end: spec, decision, code, tests, evidence, merge.

Evidence by construction

Compliance is a byproduct of shipping, not a separate phase. Every slice must clear CI gates, security scans, and acceptance tests before it merges.

Stage as DAG

Forget time-boxed sprints. A Stage is computed from the slice dependency graph — parallelism happens where the topology allows it, not where a calendar says it should.

In practice

How it works

Three moves. Every unit of work leaves an audit trail behind it.

01

Declare the slice

State the outcome, the scope, and the decision it carries — one chat, one PR. The boundary is set before an agent writes a line.

02

Agents execute

One or many agents do the work inside the slice: spec, code, tests, security scan. Humans hold the merge button and the architectural calls.

03

Evidence stays auditable

The decision record, the diff, the tests, and the approval live together in the PR. The audit trail is a byproduct, ready the moment you need it.

What sets it apart

Built for running agents at scale

Two capabilities that set the framework apart — the ones teams feel first.

The right model for every slice

Sensitive work routes to a local model automatically — explicit, on the record, compliance by construction. Not a fixed model, not opaque routing.

Model Triage

Your LLM bill, gated not surprised

Slices are sized on two axes — cost and viability — and inference spend is capped in CI like any finite resource. No invoice-day surprises.

Cost economy

Proof

Extracted from production, not theory.

SliceOps came out of building real products in regulated markets with two engineers and a fleet of Claude Code agents. The numbers below are measured by execution time — not modeled, and not padded with calendar time.

800+
Decision records in production
199M
Tokens measured in one block
13
Agents running in parallel
2.5×
Efficiency multiplier, measured

Every number is measured in production — see how →

Community

Get involved

SliceOps is open. It gets better the more people build with it and report back.

Contribute

The spec and tooling are opening under CC BY 4.0 and MIT. Open an issue, propose a principle, send a PR.

Publish your results

Run a block, measure it, write it up. Your case study becomes the next reader's training data — the framework's compounding loop.

Become a champion

Adopting SliceOps on a real team? Tell us. We're gathering early adopters to shape v1 and share what actually works.

Ship with agents. Audit by construction.

Get framework updates — the spec, the toolkit, and the Datta case study as they open.

No spam. Framework news only. Unsubscribe anytime.

Spec & toolkit

The specification and tooling are opening under permissive licenses — CC BY 4.0 for the spec, MIT for the code.

SliceOps on GitHub

Spec v1.0 — opening soon.

SliceOps is platform-agnostic — adopt it on any agent runtime. Rootblocks is the platform we build to operationalize it.