Plan → run → audit → review

aidd

A local control panel for planning AI coding work and reviewing the results.

Point it at a codebase, give it a spec, and choose the agent or model you already use. aidd builds the backlog, runs bounded agent sessions, records validation results, and flags incomplete or blocked work for review. Manage one project or several from the same local panel.

Get latest source View source

Every capability included
No account
No telemetry sent. No phone home.

Backends: Claude, Codex, local
Control: Time and iteration limits
Verification: Project checks and results
Cost: Free, every capability

aidd was built out through seven creation lanes across three languages, with real gates, published failures as well as passes, and run ledgers you can inspect.

See build results Read changelog

How it works

From codebase to reviewed changes.

aidd turns a spec into backlog items, selects eligible work, launches bounded agent sessions, and checks completion reports against feature status. Blocked work is kept visible for review.

aidd dashboard with project health, waiting approvals, and the Director queue — The aidd dashboard, captured September 8, 2026. Project health and work awaiting review in one view.

Ingest the codebase

Add application roots in Settings. Any folder with an .aidd/ contract becomes a project, including foreign stacks.

Plan the backlog

Onboarding turns a plain-language spec into dependency-ordered feature work with verifiable acceptance criteria.

Run your coding agent

Use Claude Code, Cline, Codex, Grok, OpenCode, or KiloCode, or OpenAI-compatible endpoints including OpenAI, LM Studio, and Ollama.

Check the results

The agent is instructed to run the project's required checks and report the results, including live browser verification when the acceptance criteria call for it. Review the evidence and check coverage.

Review the run

Use the run history, iteration logs, and code diffs to review what changed. Usage data is included when available. Resolve blocked work and approval requests before continuing.

Where it fits

Manage the work across sessions and projects.

aidd brings a feature backlog, repeatable workflows, run history, and review queues into one local application. Whether you need it depends on how you want to manage the work.

Use aidd when

You want a shared view of planned work and run results.

You work across repositories or reuse workflows, and want to track acceptance criteria, validation results, and work that needs approval.

Skip aidd when

You want autocomplete, a replacement IDE, or one-click unattended autonomy.

aidd is not the coding agent. If your current sessions and project tools already give you the tracking you need, it may add more process than you want.

Cloud agents

Hosted execution

Hosted agents make it easy to hand off work without keeping a local machine running. Pricing, sandbox access, and the evidence returned with a run vary by provider.

aidd: runs locally through your chosen CLI or model backend, with iteration and wall-clock limits. Usage is recorded when the backend reports it; provider charges still apply.

Agent boards

Session visibility

Kanban boards, worktree apps, and terminal managers provide useful views of active sessions and diffs. Planning depth, automated gates, and platform support depend on the product and configuration.

aidd: ties runs to a feature backlog and acceptance criteria, records validation results, and flags blocked work for review.

Spec kits

Requirements first

Spec-driven kits help turn an idea into structured requirements. How those requirements reach an agent and how results are checked depends on the kit and workflow.

aidd: connects the spec to dependency-ordered work, acceptance criteria, coding runs, and a queue for decisions that need you.

Build results

The proofs used real apps.

aidd built and repaired real applications through seven different creation lanes. The runs were there to find rough edges, file the failures, fix them, and leave a public trail.

All seven are published in full - app snapshots, run ledgers, iteration evidence, and scrubbed transcripts - in the build-proofs repository, including the lane that failed. This is first-party evidence, not a benchmark or an independent audit.

Lane	App	Stack	Result
Fresh scaffold from a spec	Habit tracker	Bun + TS + React	25/25 features passing, gate clean, roughly 2.5 h
Third-party template init	Kanban board	Vite + React	23/23 features, 10/10 headless acceptance scenarios in the gate
Full-stack template init	SMB infrastructure dashboard	Bun + Elysia + React + Drizzle	26 features passing plus one retained remediation record, 30-step gate green
Existing-codebase ingest	Flask tutorial app	Python + Flask	Ingested and audited the existing Flask app, then completed a separate security-remediation phase under its pytest gate. The initial ingest was metadata-only; remediation changed application code.
GitHub-template clone	Pantry tracker	PowerShell + Pode + htmx	26/26 features, Pester 88/88, analyzer clean
Local model only	Markdown TOC tool	Bun + TS, LM Studio, no cloud calls	0/14 features. Five coding runs finished nothing, and the completion contract refused to record success. Published exactly as it ran.
Agent-driven build	Margin calculator	Bun + Elysia + React + Drizzle	29/29 features across 48 supervised iterations, gate green at the end

Trust model

Where data goes and what limits a run.

The default posture is deliberately plain: local data, no account, explicit model backends, bounded runs, and documented write surfaces.

Local

Records stay on your machine

The runtime, panel, ledger, and project metadata live locally. Coding runs send prompts and relevant project content to the backend you choose, which may be a cloud service or a local model.

Quiet

No account, no telemetry

There is no aidd account or hosted control plane, and aidd sends no telemetry. Configured integrations and downloads you initiate can also make network requests. The write allowlist documents what lives under .aidd/ and when source can change.

Bounded

Hard limits on runs

Iteration caps, wall-clock caps, dirty-tree guards, and parked decisions stop a run outright; cost and token budgets are warn-only, reported against the run rather than cutting it off mid-edit.

Open

Security docs in public

Review SECURITY, PRIVACY, and SUPPORT.

Included under FSL

Every capability is included.

Run aidd supervised or unattended, on one project or across a fleet. Every capability ships together under FSL. You provide access to the AI providers you choose.

aidd

One product

Everything aidd does ships together, for any number of projects and machines.

Runs and gates
All backends
Audits
Recipes and skills
Director cycles
Scheduled fleet automation
Continuous follow-ups
Telegram bridge
MCP server

Source-available under FSL-1.1-ALv2, converting to Apache 2.0 two years after each release.

Install from source

Project state: Your local .aidd/ metadata
Control plane: Your machine
aidd account: None
Telemetry sent: None
Model access: Your configured provider or agent CLI

Get it

Four commands. Nothing to sign up for.

Windows 11 + PowerShell 7 is the current smoke-tested path. Install Git and Bun 1.4.3 or newer, then run aidd directly from its source checkout.

Source install

git clone https://github.com/NomadicDaddy/aidd.git C:\Tools\aidd
Set-Location C:\Tools\aidd
bun install
bun run start:web

# Open http://127.0.0.1:3210

aidd ships under FSL-1.1-ALv2: available for any use except making it available to others as a competing commercial product or service, and each version automatically becomes Apache 2.0 two years after release. The .aidd/ contract format is an open spec anyone may implement. Spernakit, the companion full-stack template, is MIT.