git.agentic
Three projects · all open source Apache 2.0 Rust · TypeScript · Python

The tooling under software assumes a human in the loop.

AI agents broke that assumption. git.agentic builds open-source version control, source control, and software supply-chain security for software written and operated by agents.

git revert     assumes code is the only thing that changed   → git.agentic
npm install    assumes a human read what it fetched          → Sentinel
your checkout  assumes one pair of hands                     → src-control

git.agentic

Git for agent behavior.

An agent's behavior is determined by six things — code, prompts, tools, model, memory, schema — and git revert only knows about one of them. git.agentic snapshots the whole tuple atomically and rolls all six dimensions back coherently, including reverse schema migrations and memory state.

$ agentic rollback v0.7
  ✓ Schema reverted          in 0.4s
  ✓ Memory restored          in 2.1s
  ✓ Prompts restored         in 0.0s
  ✓ HEAD now at i7j8k9l (rollback of v0.8 → v0.7)

Rust core · Python SDK · open-source MVP complete · production validation in progress

Full page →   GitHub →

src-control

Version control built for fleets of agents.

A checkout assumes one developer, one directory, one disk. Agents want N parallel worktrees, now, and gone without a trace afterwards. src-control forks copy-on-write worktrees entirely in RAM — checkout to disk only when you ask — with native committed secrets, signed commits, sealed agent-session transcripts, private branches with atomic publish, bidirectional Git sync (local or GitHub), and a native read-only desktop browser for refs, snapshot DAGs, provenance, public trees, and diffs.

# fork 4 parallel in-RAM worktrees off one snapshot
$ sc demo --agents 4

# agent workspaces: run a command across N forks, harvest results
$ sc work --agents 4 -- cargo test

Rust · Tauri v2 · TypeScript · Phases 1–35 built

Full page →   GitHub →

Sentinel

An agent-auditable security layer and registry for npm.

Agents run npm install and execute code nobody read. Sentinel is an auditing proxy in front of registry.npmjs.org that grew into a first-class registry: every tarball is scored by a deterministic engine and carries a verdict before install-time code runs, a deny-by-default sandbox contains what does run, and claimed names get verified ownership and time-locked retraction. The agent can request; only a human grants.

  color-stream@1.4.1
  ────────────────────────────────────────
  install    ⚠ runs lifecycle scripts
  score      ░░░░░░░░░░ 0/100
  verdict    BLOCK
  critical [install-scripts]  postinstall decodes an encoded blob
  critical [secret-exfil]     reads ~/.npmrc, AWS credentials…

  HTTP 403  x-sentinel-verdict: block

TypeScript · security layer + registry built · pre-1.0 · v0.1.0-alpha.1 on npm

Full page →   GitHub →

Technical guides

Clear answers about versioning and rolling back AI agents.

Read all guides →