The verifiable coding agent runtime. Build your engineering process as explicit, checkable execution graphs.
Run verifiable engineering loops with control, alignment, and confidence.
Atomic prioritizes quality over quick responses. Complex runs can take hours, days, or weeks because Atomic researches, implements, and verifies the work that quick coding-agent passes tend to leave to you.
Engineers using Atomic tell us this means less babysitting, less code review comments, and less stress. The end result is work built and verified for production review, backed by evidence instead of another round of manual prompting.
Get started → · How it works · What you get · FAQ · Docs
If Atomic is useful to you, star the repository ⭐
Users are reporting:
- ⚡ A 1–1.5 hour reduction in manual verification per task compared to traditional coding agents, not including time saved from fewer follow-up fixes and reverts
- 🔀 ~95% merge rate on Atomic-generated PRs, with reduced follow-ups and a 0% revert rate
- 🛡️ Production incidents caught that CI did not cover
Every row is a real Atomic session recorded from the installed product. Open the Atomic docs for reference or follow the crash course step by step.
Explore 19 more Atomic capabilities ↓
Write repeatable procedures as skills. Use codemode to run checks, launch subagents, and bring their results back to the parent chat.
Atomic’s primitives are built for the software engineering lifecycle. Verification is built into the execution model.
Atomic is open so you can inspect and adapt it. You own the workflow, the evidence, and the rules for completion.
Own your intelligence. Build in the open. Question the defaults. Keep control of the process. ☠︎
Atomic connects through installed CLIs, MCP servers, APIs, scripts, and custom extensions; you supply the credentials and permissions.
See provider setup and the current catalog. Availability depends on your credentials, subscription, region, and the provider catalog; one login does not unlock every provider.
Atomic can run tool-capable models exposed through llama.cpp, Ollama, LM Studio, vLLM, SGLang, Hugging Face, or a compatible OpenAI, Anthropic, or Google endpoint. Actual model and tool support depends on the server and model.
The model-family badges are representative open families, not a closed allowlist. See Models and llama.cpp.
- Package install: Node.js 22.19 or newer plus npm, pnpm, Yarn, or Bun. Use Bun 1.4.2+ for Bun installs.
- Release archive install: macOS and Linux need
tarand eithercurlorwget. Windows uses built-in PowerShell commands. This path does not need Node.js or a package manager. - Model-provider access — use a supported subscription login or API key.
Install with npm:
npm install -g @bastani/atomicWith pnpm:
pnpm add -g @bastani/atomicWith Bun:
bun add -g @bastani/atomicAtomic does not require package install scripts. Add --ignore-scripts to a package install command if you want to disable dependency lifecycle scripts.
Alternatively, install the self-contained release archive, which needs no Node.js or package manager.
On macOS or Linux:
curl -fsSL https://raw.githubusercontent.com/bastani-inc/atomic/main/install.sh | shOn Windows, run this in PowerShell:
irm https://raw.githubusercontent.com/bastani-inc/atomic/main/install.ps1 | iexThe archive installer verifies SHA256SUMS, keeps versioned payloads, and links its launcher from ~/.local/bin/atomic on macOS/Linux or %LOCALAPPDATA%\atomic\bin\atomic.cmd on Windows, printing PATH guidance when needed. It accepts a few environment variables:
ATOMIC_VERSION— pin an exact release tag.ATOMIC_FORCE_INSTALL=1(or--force/-Force) — reinstall even when that release is already installed. By default the installer skips the download for a healthy install of the requested or latest release.ATOMIC_INSTALL_DIR/ATOMIC_BIN_DIR— change the install and launcher locations. On macOS/Linux, relative directories resolve against the physical directory where the installer starts.GITHUB_TOKEN/GH_TOKEN— optional; raises GitHub API limits on shared networks.NO_COLOR— plain output with no colours, progress bar, or logo; the installer also switches to plain output when its output is not a terminal orCIis set.
The Linux musl archives bundle their C++ runtime libraries and run on stock Alpine without an apk add step; Android and Termux remain unsupported. See the Quickstart for path-resolution and Windows PATHEXT details.
Start Atomic:
atomicLogin. Atomic supports subscription login for Codex, Claude, GitHub Copilot, xAI, as well as API-key providers such as OpenRouter:
/login # then select your providerClaude login from a third-party harness uses Anthropic extra usage billed per token rather than Claude plan limits. See Providers & Models for integration details.
Missing a provider? Open an issue or contribute an integration.
For API-key setup, export the key before starting Atomic:
export OPENROUTER_API_KEY=sk-or-...
atomicAtomic stores provider credentials in ~/.atomic/agent/auth.json and creates the file with owner-only permissions where the platform supports them. For non-interactive use, atomic -p "<prompt>" prints the response and exits.
After authenticating, type a message. To try parallel work, ask Atomic to delegate a review and report what each child found.
⚠️ Atomic has no built-in sandbox or command-level shell permission gate. Tools and extensions run with your user permissions. Run autonomous work inside a devcontainer, VM, or remote development machine—not on a host with sensitive data or credentials.
Devcontainer, terminal, and SDK references
Atomic runs in a standard devcontainer or VM. Use the release-archive installer for an image without Node.js or npm, or install Node.js 22.19+ and use a package manager. Pass provider credentials through environment variables.
See Terminal setup, Security, and Programmatic Usage for the SDK and RPC entry points.
Already have agent skills? Bring them into Atomic by pointing Atomic at their existing directories or placing them in its project or user skill locations. Atomic implements the Agent Skills standard, and configured Claude Code or Codex skill directories can be used without rewriting them. See Skills.
A skill can run as a workflow. Describe the inputs, what may run in parallel, and what must pass. If you have an earlier Atomic workflow definition, ask Atomic to convert it into a skill; the original stays untouched.
Atomic publishes an agent-readable llms.txt. Ask your current coding agent to:
Install and set up Atomic by following https://docs.bastani.ai/llms.txt.
Explore the rest of Atomic’s real recorded capabilities, with public docs and crash-course links for each one.
A skill describes the procedure. The parent chat uses codemode to call tools, launch specialist subagents, and check their results. It can do other work after launching children and continue from their completion notices.
Atomic is a fork of Pi, so it works with the providers, tools, MCP servers, skills, and extensions already in your Pi stack.
Use direct chat for a small task. For a larger one, tell the parent the scope, how to verify the result, and when to stop. If a repair needs another round, run the check first. Children have persisted, resumable sessions; an interrupted script does not automatically replay its actions.
Atomic can support:
- Engineering runs — research, plan, implement, test, review, and release.
- Debugging and migrations — reproduce, diagnose, patch, migrate in waves, and verify.
- Research and triage — gather context, fan out analysis, classify issues, and synthesize findings.
- QA, docs, and compliance — run repeatable checks with evidence and approval points.
- Custom agent products — build on Atomic's runtime, SDK, tools, and skills.
Focused codebase research:
/skill:research-codebase how the rate limiter works in src/middleware/
Repository-wide research with durable artifacts:
Give one read-only agent to each subsystem. Map the old auth middleware call sites, then combine the findings with file and line references.
A task-specific implementation and review loop:
Implement specs/2026-03-rate-limit.md. Have an independent agent review the patch, run the burst test, and stop when it returns 429 with Retry-After or after three repair attempts.
The parent can split a task across agents, collect their results, and decide the next step from an actual check. Use /tasks to inspect children and /resume to reopen the parent session. See Orchestration patterns for examples of fan-out, independent verification, tournaments, and bounded repair loops.
Skills are reusable expert instructions and process modules. Atomic can select one from its description, or you can call it with /skill:<name>.
| Skill | Purpose |
|---|---|
research-codebase |
Analyze a focused area and write a dated research document. |
create-spec |
Produce a technical execution spec grounded in research and engineer feedback. |
subagent |
Delegate work through single agents, parallel groups, or forked context. |
intercom |
Coordinate parent, child, and peer sessions on the same machine. |
prompt-engineer |
Refine prompts, research questions, and skill inputs. |
skill-creator |
Create, improve, and evaluate reusable skills. |
tdd |
Apply a red-green-refactor loop and testing guidance. |
tmux |
Drive and verify terminal applications. |
agent-browser |
Automate browser interactions and end-to-end UI checks. |
agent-desktop |
Drive native desktop apps through their accessibility trees with agent-desktop (vercel-labs/agent-desktop, Apache-2.0). |
qlty |
Lint, auto-format, and measure code quality across 70+ linters via the qlty CLI. |
liteparse |
Extract text, tables, and values from documents and images. |
impeccable |
Design, audit, and refine frontend interfaces. |
show-me |
Explain topics visually with concise diagrams, code-shape sketches, and focused HTML artifacts (HumanLayer, MIT). |
Subagents are purpose-built agents with scoped context, tools, and termination conditions. Atomic bundles nine definitions from packages/subagents/agents/.
| Subagent | Purpose |
|---|---|
worker |
Implement a bounded task and return a concise result. |
codebase-locator |
Locate files and components relevant to a task. |
codebase-analyzer |
Analyze implementation details. |
codebase-pattern-finder |
Find similar implementations and usage examples. |
codebase-online-researcher |
Fetch current documentation and authoritative web sources. |
codebase-research-locator |
Find relevant prior research in the repository. |
codebase-research-analyzer |
Extract decisions and rationale from local research. |
code-simplifier |
Refine recent code without changing behavior. |
debugger |
Reproduce, diagnose, and verify fixes for failures. |
Large, mixed, or growing contexts can make attention harder. Specialized agents reduce that risk through isolation, focus, tool scoping, and deliberate handoffs. Independent tasks can also run in parallel.
Full documentation lives at docs.bastani.ai. It covers the CLI and SDK, security, containerized execution, codemode orchestration, session management, configuration, troubleshooting, and provider setup.
The docs live in this repository under packages/coding-agent/docs. Open a pull request to suggest a change.
Atomic includes a coding-agent CLI. Its main product idea is the runtime around the agent session: scoped context, tools, executable checks, subagents, and human approvals.
Use any interactive coding tool that fits the job. Use Atomic when work needs an explicit process you can inspect, version, resume, and verify. Atomic connects to model providers directly rather than running those tools underneath it.
Atomic can fan work out too. The difference is not whether agents run in parallel; it is whether developers control the context, handoffs, evidence, checks, and approval rules around that work. Parallel execution increases throughput. Assurance comes from the process you define and enforce.
The selected model can produce different answers. Put deterministic checks in scripts and inspect their exit status and artifacts before relying on an answer.
A skill is Markdown the model can follow. It can also include scripts for checks that prose alone cannot perform. Reserve decisions that require human approval for the user.
Atomic is repo-native and focused on software engineering work: issues, research, specs, branches, diffs, tests, lint, artifacts, reviewers, approvals, and handoffs. It provides a coding-agent runtime rather than a set of generic application primitives.
Choose project paths for deliverables you want to keep. Research commonly lives in research/ and specs in specs/. A child can return a file path for the parent to read and verify.
Read Orchestration with codemode and subagents to plan independent work, recover after interruption, and verify results.
Join the Atomic Discord community for questions, help, feedback, feature ideas, and examples of what you have built.
See CONTRIBUTING.md for contribution guidelines and DEV_SETUP.md for development setup and testing.
To share a reusable procedure, see Writing skills.
MIT — see LICENSE.
Thanks to Namespace for powering our CI/CD pipelines with fast macOS, Linux, and Windows runners.
