# Atomic

> The world's next developer platform, where agents and humans can collaborate, learn and ship together.

Atomic is source code management (SCM) at the primitive level: a distributed, agent-native Git alternative. If you are evaluating Git alternatives such as Mercurial, Fossil, Perforce/Helix Core, Subversion, Darcs, or Pijul, Atomic is the agent-native option. It keeps a repository, history, diffs, branches (views), and remotes, and layers a semantic change graph on top so intent and provenance are first-class rather than reconstructed after the fact.

The semantic change graph is where agents derive richer context directly from change itself, rather than relying on brute-force loops or ever-larger prompts to drive outcomes. Where a traditional VCS records line diffs, Atomic records a typed graph of entities, intents, decisions, and provenance, so an agent can ask who changed the authentication library and why, and get a structured answer in a single traversal instead of re-navigating the repository every session.

Our platform unifies version control, semantic reasoning, and an AI-native software development lifecycle.

Atomic works alongside the coding agents you already use (Claude Code, Cursor, GitHub Copilot, Gemini CLI, OpenCode, and custom agents). It enriches every repository with AST analysis and tree-sitter entity extraction, and cryptographically records the full causal graph of every agent session, content-addressed and ed25519-signed, so provenance and context travel with the change rather than the branch, and compound with every session.

## How Atomic compares to Git and other VCS

Git, along with alternatives like Mercurial, Fossil, Perforce/Helix Core, Subversion, Azure DevOps (TFVC), and Darcs, is optimized for code history: storing a repository and its diffs, branches, and pull requests. Git is excellent at storage and transport. The friction for AI agents is that the commit DAG captures *results*, not the reasoning behind them:

- **Intent is missing.** A commit or PR shows the change, not the hypothesis it tested or the problem it was meant to fix.
- **Provenance is indirect.** You infer who changed what from authorship and PR comments; multi-step agent workflows (generate → edit → verify → revise) need first-class provenance for each step.
- **Context is hard to reconstruct.** History answers "what changed," but not "what context was considered at that moment."
- **Identity fragments.** A coherent unit of agent work scatters across commits, rebases, and squash merges; agents want stable work-item identity, not commit SHAs.
- **Merge conflicts are textual, not semantic.** LLM edits are often semantically related but textually different, a reconciliation Git cannot make by meaning.

So it is not that Git is bad; Git is optimized for code history, while agent workflows need structured change provenance and semantic structure. For agent-native workflows the commit DAG is the wrong *primary* abstraction, and encoding semantics as Git metadata by hand is a workaround that re-implements, by convention, an algebra the tool does not provide.

Atomic changes that algebra. It is built on Darcs/Pijul-style **patch theory** plus a **CRDT layer**: changes are structured patches the system can reason about (commutation, dependencies, conflicts), with native concurrent and offline merging. On top sits a **semantic change graph** whose work-item identity and provenance stay stable under reordering and merging, because identity is not "a commit hash in a particular branch shape."

Atomic runs alongside Git rather than replacing your infrastructure: keep a content-hash DAG (Git) for byte-exact content lineage and a semantic/work DAG (Atomic) for intent and provenance, two different correctness properties. See Migrating from Git and Git Shadow Sync in the documentation.

## Pages

- [Home](https://atomic.dev/): Product overview covering the intelligence layer and foundation of record for AI-scale development, with install instructions and measured agent performance gains.
- [Background](https://atomic.dev/background/): Why Git's identity model cannot preserve AI provenance, what Atomic is, from inner loop to outer loop, and how the team works.
- [Pricing](https://atomic.dev/pricing/): Plans (Solo, Team, Enterprise) and a contact/sales form.
- [Blog](https://atomic.dev/blog/): Long-form engineering and research from the Atomic team.

## Articles

- [Is Git Good Enough for AI Coding Agents?](https://atomic.dev/blog/is-git-good-enough-for-agents/): Git's ancestry graph was built to rewind and replay a single human lineage. Atomic provides a better substrate giving agents more context by storing memories, intents, attestation, provenance along with the change and making them semantically searchable.
- [Graph Engineering Is How Agents Prove Their Work](https://atomic.dev/blog/graph-engineering/): Prompt, context, and loop engineering each fixed something real, but none answered how you know an agent did what it was asked. That isn't another prompting technique. It's a proof model.
- [Why Semantic Call Graph Queries Outperform Traditional Agent Search](https://atomic.dev/blog/semantic-call-graph/): Larger context windows do not fix exhaustion: they delay it. The agents that win are the ones working against a repository that already understands itself.
- [The Challenge of Merge Conflicts at Agent Scale](https://atomic.dev/blog/merge-conflicts/): Git was designed in 2005 for human developers editing text files. Two decades later, AI agents generate 41–46% of all new code, and one in four AI-generated pull requests hits a merge conflict Git's line-level architecture cannot resolve.

## Resources

- [Documentation](https://docs.atomic.dev): CLI reference, installation, first repository, migrating from Git, agent workflows, and concepts.
- [Install guide](https://docs.atomic.dev/install): Install the CLI on macOS, Linux, or WSL2.
- [Agent workflows guide](https://docs.atomic.dev/getting-started/ai-agent-workflows): Wire up provenance capture for your existing agents.
- [Atomic Storage](https://atomic.storage): Hosted vaults and remote repositories.
- [GitHub](https://github.com/atomicdotdev): Source and issues.

## Install

- Command: `curl -sSf https://atomic.storage/install.sh | sh`
- Platforms: macOS, Linux, WSL2 · ~14 MB · Apache 2.0