- 1Plan, approved
Large tasks start with a plan, and nothing changes until someone approves it.
- 2Work, sandboxed
Every command runs in a sandbox with a network allowlist, with your keys out of reach.
- 3Checks, measured
It opens what it built in a real browser, measures it, and runs your tests.
- 4Record, kept
The plan, the session, the commits and the checks stay with the project.
the product
One engine that plans, writes, checks and finishes the work.
AdultCoder is the coding agent behind every AdultCoder surface. It runs on open-weight models on your own hardware, or on hosted models with your own API keys, and it does not call a task done until the checks say so.
A shell in your terminal
The conversation lives in your terminal's own scrollback, with the prompt and a status line at the bottom. /plan, /budget, /unattended and the rest are a slash away, and the Mac menu bar app sets up keys, model servers and defaults.
Your models or your keys
Ollama, llama.cpp, vLLM or LM Studio on your own hardware; Anthropic, OpenAI, DeepSeek and Qwen with your own keys. A default, a fast and a quality model, chosen per project.
Checks it actually runs
Browser reviews that measure overflow and the phone menu, your own test suite, and a plain account of anything still broken.
what you see
A status line that tells the truth.
Long work is easier to trust when you can see it. AdultCoder says what it is doing, what needs you, and what it has spent.
- AProgress you can read
What it is working on, what waits for your approval, and the spend against your budget.
- BWhen a server is down
The status counts down to the next try. Left unattended, it keeps trying until the server answers.
what the engine does
- 1Long work that finishes
Long conversations are compacted without losing the task, and a loop stops when it is done, stuck, or over budget.
- 2Sessions that outlive a model
Any session resumes on another model, so a model that went away never strands the work.
- 3Storage that stays with the project
Sessions, memory, embeddings and permissions live with the project, on your machines.
- 4Budgets that end in finished work
A task gets a budget in dollars and minutes, a session one in dollars. It works until the job is done, stops cleanly at the limit and says what it spent; raise the budget and carry on.
- 5Proprietary and native, not a web app
Our own engine, built in-house: one Rust binary with native apps, so it starts in a hundredth of a second and stays light on large repositories.
We did not build a demo. We built the engine serious teams run on.
Pro-grade results on models you run yourself, hosted models only on your terms, and controls that hold when nobody is watching. And the rest of a product when you need it: images and app icons, voice-overs and music, video clips and finished films, under spending limits you set.
private walkthrough
See it work on the kind of project your team builds.
We will show the engine, the controls and the record it keeps, in one session.