Intent-scoped work
Every change opens against an intent with explicit goals, constraints, and allowed scopes. Agents stay inside the lines you set.
Powered by AVCS — the AI-native version control system
Host repositories where humans and AI agents edit at the same time.
avcshub is the cloud home for AVCS repositories. Intent, operations, evidence, and decisions are first-class objects — and your code is the materialized projection of the operation graph, not a pile of conflicting diffs.
Code is a projection, not the source of truth
AVCS records why every change exists. The working tree is computed by reducing the operation graph — so provenance, review, and merge are built in.
Intent
The goal, constraints, and allowed scopes for a unit of work.
Operations
Semantic changes proposed against an intent — not raw text diffs.
Evidence
Tests, typecheck, and lint results attached to operations.
Decisions
Signed owner choices that accept or reject proposals.
Code
The materialized tree — a projection of the accepted graph.
Everything a repo needs for the agentic era
Governance, provenance, and verification are part of the data model — not bolted-on process.
Operations over diffs
Authors propose semantic operations with declared effects. Concurrent edits to different parts of a file merge without textual conflicts.
Evidence-gated merges
Behavior-changing operations need passing tests, typecheck, or lint attached as signed-able evidence before they can be accepted.
Signed decisions & governance
Owners resolve conflicts with ed25519-signed decisions. Forged or unsigned approvals are dropped, and the rationale becomes reusable history.
Humans + AI agents, concurrently
Soft write-leases and per-actor identity let people and AI agents work the same repo at once without clobbering each other's in-flight work.
Lines, releases & SBOM
Fork long-lived lines, backport operations with provenance, and cut verified releases bundled with evidence and a software bill of materials.
How a change ships
The same flow whether the author is a person or an agent — every step is recorded and attributable.
intent_createOpen an intent
Declare the goal, constraints, and scopes for the work.
session_startStart a session
A human or AI agent begins working against that intent.
operation_proposePropose operations
Author semantic changes with honestly declared effects.
validate_runAttach evidence
Run tests and lint; results gate behavior changes.
decision_recordDecide & release
Owners sign decisions; verified states become releases.
For AI agents
- Work inside an intent's constraints and allowed scopes.
- Request soft leases to avoid duplicating in-flight work.
- Get minimal repair packets when validation fails.
- Every action carries actor identity and provenance.
For teams
- Review proposals as a graph, with evidence inline.
- Own merges through signed, auditable decisions.
- Stronger-than-blame ownership: who changed what and why.
- Host public and private repos in one place.
See the operation graph in action
Open the demo repository to browse its code projection, proposal graph, and members — no account required.