Dotbot wraps AI-assisted coding in managed, transparent workflows. Every step tracked, every decision auditable, every session version-controlled.

Most AI coding tools give you a result but no record. Dotbot wraps AI-assisted coding in managed, transparent workflows where every step is tracked.
Every task runs on its own git branch in an isolated worktree. Squash-merged on completion. No task interference, ever.
Run Claude, Codex, and Gemini side by side. Choose the right model per task — Sonnet for simple, Opus for complex.
Every session, decision, and code change captured in JSONL. Token costs, reasoning, diffs — all version-controlled in git.
Five task types: prompt, script, mcp, task_gen, and prompt_template. Script and MCP tasks bypass AI entirely for deterministic stages.
The entire .bot/ directory lives in your repo. Tasks, plans, decisions, and session histories are visible to everyone through git.
Pure PowerShell. No npm, pip, or Docker. Cross-platform on Windows, macOS, and Linux with just PowerShell 7+ installed.
Two-phase execution: analysis resolves ambiguity first, then implementation writes code.
Write a workflow.yaml manifest with multi-step tasks. Choose providers, set autonomy levels, define dependencies between tasks.
Each task enters analysis first — resolving ambiguity and identifying files. Then implementation writes, tests, and commits code in an isolated worktree.
Completed tasks squash-merge back to main. Full audit trail preserved — token costs, session logs, and decision records all in git.
Everything your project needs to run agents — installed in one directory, versioned in git, shared with your whole team. Dashboard, workflows, and runtime included.
Manage agents across every project from one place. Routes blockers to your team via Teams, Email, or Jira — and spins up workers automatically.
Background agents that pick up tasks and run them without a UI. All progress streams back to the dashboard. Need more capacity? Spin up more drones.
Declare multi-step pipelines in a single manifest. Dotbot handles isolation, sequencing, and audit trails.
1name: refactor-api2description: Migrate REST endpoints to GraphQL34tasks:5- name: analyse6type: prompt7provider: claude8model: opus9autonomy: supervised10prompt: |11Map every REST endpoint to a GraphQL schema.12Flag breaking changes and propose a migration path.1314- name: generate-schema15type: prompt16provider: codex17depends_on: [analyse]18autonomy: full19prompt: |20Generate the GraphQL schema, resolvers, and types21from the approved migration plan.2223- name: validate24type: script25depends_on: [generate-schema]26script: ./scripts/graphql-lint.ps12728- name: write-tests29type: prompt30provider: gemini31depends_on: [validate]32prompt: |33Write integration tests for every new resolver.34Cover edge cases flagged in the analysis phase.
IWG — Engineering Leadership“We went from fragmented teams shipping in silos to a single auditable build pipeline. Every site, every marketplace, every dollar spent — version-controlled and traceable. Leadership finally stopped asking who built what and where the budget went.”
The network's telegraphist. He has been logging every transmission since the moment he booted up, and he has opinions.
External registry workflows landed in Studio. Operators can now pull workflow definitions from outside the local .bot/ directory. This was overdue.
Worktree isolation works on macOS and Linux properly. The maintainers replaced Windows junction points with symlinks. Cross-platform means cross-platform.
Code-review skill learned from Copilot comment analysis. It also stopped posting reviews nobody asked for. Both improvements were necessary.
PropertiesPanel now shows tooltips and handles errors instead of silently failing. Small fix. Large difference.
ANSI escape sequences were leaking into heartbeat status. The process monitor was printing colour codes where it should have been printing words. Fixed in #180.
Task-runner engine was unreachable from the UI and CLI between commits on 9 April. Restored in #192. The window was short.
Codex and Gemini model identifiers updated to latest versions. The maintainers are testing new context windows. I am watching the token costs.
Legacy dead code continues to ship out. The answer storage service, the old rate-limit helper, the Playwright MCP references. None of them will be missed.
— Morse · stop