Frequently asked questions

Short answers to the questions a first-time visitor asks, each with a link to the full guide. If your question isn't here, ask in the app or use the contact page. See Getting help.

What is it?

What is DevSpec, in plain terms?

DevSpec is where your team, your knowledge and your agents build software together, made for AI-native development from the ground up. You talk a change through in a session, the decision becomes tasks with acceptance criteria, your own agent builds them, and DevSpec follows the work through the deploy to the check that it works. See What is DevSpec, Why DevSpec and the Quickstart.

Who is it for?

Teams building real software with AI agents: founders, executives and product leaders who want to shape the product directly and see honest progress; developers who want their coding agents grounded in the project's codebase, schema, logs and past decisions; non-technical teammates in product, design or operations who want to contribute in plain words and run safe experiments in the same room as engineers; and agencies or consultancies working alongside clients with clear role boundaries. Solo builders get the same persistent project memory and verification loop as they grow. See Why DevSpec and What is DevSpec.

How is this different from just using an AI agent on its own?

An agent on its own starts every conversation blind. Nothing it learns survives the session unless you carry it over by hand. DevSpec keeps the shared record around the agent: decisions and conventions in memory, the discussion that led to the work, and verification of what shipped. Your agent arrives informed instead of starting from zero. See How work flows through DevSpec and Memories.

Do I need to be a developer to get value from it?

No. You join sessions in the browser, describe what you want in plain words, answer questions when work needs your call, and see how your input changed what got built. Engineers handle the code; you handle the decisions only you can make. See Sessions.

How does this help my team work faster while staying in control?

People and agents share one room, so nothing waits on handoffs. Control stays with people: a person always decides when work starts, agents stop at Implemented and never mark work Done, and verification shows what was checked and where your judgement is still needed. See How work flows through DevSpec.

Agents and harnesses

Which agents does DevSpec work with?

Pi, Claude Code, and Antigravity CLI (agy). All three run on your machine, use your own model subscription or API key, and get the same DevSpec features. You can connect any of them, on as many computers as you like. See Connect your agent.

Why do you recommend Pi?

Pi is open source and works with whatever model subscription or API key you already have, and from the DevSpec web app you can stop it mid-task and change its model. The one exception is a Claude subscription, which only works in Claude Code, so if you pay for one, connect Claude Code. See Connect your agent.

Do I have to switch from the agent I already use?

No. If you use Pi, Claude Code or Antigravity CLI (agy), connect what you have. Setup walks you through it, and you can talk to Dev and plan work before connecting anything. See Connect your agent and the Quickstart.

Do I need my own AI subscription or API key for my agent?

Yes. Your agent's model runs on your own subscription or API key, and its provider bills you for it. DevSpec doesn't charge for connecting an agent, for an agent waiting, for handing it work, or for posting in a session. See What agents cost and Billing, usage and limits.

Does DevSpec pay for my agent's model usage, or do I?

You do, through your own provider. DevSpec charges your organization's balance only for the work DevSpec itself does, such as Dev's replies and background work for your projects. See Billing, usage and limits.

Can I use different models?

Pi works with the model subscription or API key you already have, and you can change its model from DevSpec in the browser. Claude Code uses Claude, and Antigravity CLI uses the models configured in your Antigravity account. See Connect your agent.

Can I run several agents at once?

Yes. Start Pi, Claude Code or Antigravity CLI in several terminals in the project's folder and each one appears separately on the Agents page with its own codename. Handing over a parallel set of items keeps them from colliding: items that touch the same files stay in one batch, and if two agents are asked for the same item, the second skips it and says so. See Working with your agents.

Can I get two agents to compare approaches or review each other's work?

Yes. Bring both agents into one session and ask them. People do this for second opinions before anything gets built, and for reviews after. Each reply is labeled with the agent's name and owner, so you can see who said what. See Working with your agents and Sessions.

Can I use a teammate's agent?

When the agent is idle, yes, if its owner allows it. New agents default to Only you, and the owner can open an agent (or their default for new agents) to Anyone on the project or Chosen people. The controls stay with the owner: only they can end it, stop it or change its model, and work done at your request is recorded under your name. See Working with your agents.

Does my agent run on my computer or in the cloud?

On your computer. That's where it reads your files and runs commands, so keep the computer awake and online while it's working. You steer it from anywhere through the web app, including your phone. See Working with your agents.

Can I control what my agent is allowed to do?

Yes, in layers. Changes an agent makes in DevSpec use the project permissions of the person asking it. Tokens come as read and write or read only. Projects set protected paths it must never touch, plus execution rules such as checks to run before pushing. And each agent has a setting for who may talk to it: only you, anyone on the project, or chosen people. See Connect your agent and Rules and instructions.

Can I start or steer my agent from my phone or the web?

You can steer from both. From the web app or your phone's browser you can see every agent, hand over items, and message an agent in a session. Starting an agent and copying terminal commands are for computers, so those controls don't appear on phones. See Working with your agents.

Can I trust it?

What access does DevSpec need to my code, and can it change anything without me?

DevSpec reads your tracked repositories to index and search them. Dev, the in-session assistant, reads code but never changes it or starts a coding agent. You choose what your agents work on, either by asking directly or by enabling an automation. Agents commit and push work branches, and out of the box those merge into your tracking branch, so point tracking at a branch your team doesn't push to directly, or switch off auto-merge to review each branch first. See How work flows through DevSpec and Repositories and GitHub.

Is my code or conversation used to train AI models?

No. Your data is encrypted at rest and in transit, and your content is never used to train AI models by DevSpec, and DevSpec never sells it. The providers that process your content to produce answers are named in the privacy policy. See Your settings.

Who can see my project, and where is my data kept?

Only people in your organization, and only as far as the project's access setting allows: available to everyone, or invite only. Each session also has its own visibility. In privacy settings you can export your data and close your account. See Organizations, members and roles and Your settings.

What does it cost?

How does pricing work?

Each organization has one prepaid balance in dollars. DevSpec charges it per piece of work, with no minimum: Dev's replies, voice, web searches, and background work for your projects such as indexing code and reading commits and deploys. There are no seats, subscriptions, plans or tiers. See Billing, usage and limits.

Can I try it for free, and do I need a card?

Your first organization starts with a $15 balance, so you can try DevSpec before adding any money. Organization Owners and Admins save a card during setup, and nothing is charged: the card is only used if you add funds or turn on auto top-up. People you invite don't need a card. The in-app help assistant is always free. See Billing, usage and limits.

Do I pay per team member?

No. Invite as many teammates as you like; your organization pays only for the work DevSpec does, from one prepaid balance. See Billing, usage and limits.

How do I keep spending under control?

Owners and Admins can switch Dev, reading aloud and voice input on or off per person, give each person an allowance that covers all three, and set a monthly limit for each project or pause it, all under usage and limits. Auto top-up always has a monthly limit you choose. See every charge under transactions and where the money went under spend.

What happens if we run out of money?

When the balance reaches $0, the work that costs money pauses: Dev, reading aloud, voice input and background work such as indexing. Everyone can still read, post in sessions and manage their work, and connected agents keep working. If no funds are added within 30 days, the organization becomes read-only: you can still sign in, read everything and export, and nothing is deleted. Adding funds restarts everything at once. See Billing, usage and limits. See Billing, usage and limits and Troubleshooting.

Using it day to day

Can my whole team work in it together, and can I control who sees what?

Yes. Sessions are shared rooms with their own visibility, projects are open to everyone or invite only, and roles decide who can change settled records. Teammates can talk to your idle agents, and every reply and change is labeled with who made it. See Sessions and Organizations, members and roles.

Does it remember decisions and context between conversations?

Yes. Settled decisions live in the project's memory, past sessions stay searchable, and Dev searches both when a topic comes up. The Memory section below explains how that record works. See Memories and Sessions.

How does a change go from chat to shipped?

Talk it through in a session, record the work as tasks with acceptance criteria, hand them to an agent, and DevSpec links the deploy back to the items and checks the result. Verification shows what was tested and observed, and the item reaches Done when its criteria are met. See How work flows through DevSpec.

Connecting hosting, databases and logs

Why does DevSpec ask me to connect Vercel, Coolify, my database or my logs?

So it can understand your project, not just your code. Deployments show whether shipped work actually landed. The database shows your real tables. Logs show what broke in production. DevSpec connects to these so it understands the whole picture, and your agent gets that understanding through DevSpec without needing its own access. None is needed to start; connect them when a conversation calls for it. See Delivery and testing, Database and migrations and Log providers.

Can DevSpec deploy, change my database or touch my live app through them?

DevSpec watches deploys; it doesn't deploy for you. When a deploy fails, it records the failure and notifies the right people. You can ask your agent to investigate and fix it, or enable a failed-deploy automation to do that with the permissions you choose. It never writes to your database: for a connected database it reads the structure, the migration history and advisor results, never your application data. See Delivery and testing and Database and migrations.

What does DevSpec read, and does it keep a copy of my data or logs?

From deployments it reads deploy status and results. From your database it reads the structure of your public schema, migration history and advisor results, never application data. From logs it searches the datasets you choose when Dev needs them. It keeps the record any working system keeps: deploy history, schema knowledge and memories, each visible in the app. See Delivery and testing, Database and migrations and Log providers.

How does it use what it learns? Does it become memories?

When a conversation settles something worth keeping, Dev offers to record it as a memory, and your agents can record durable findings too. Deploy results link back to the tasks that shipped them. Day to day learning becomes project knowledge instead of vanishing with the session. See Memories and Delivery and testing.

Do I have to connect them? What do I miss out on if I don't?

No. Without them you still get sessions, tasks, agents and memories. What you miss is grounding: no proof that shipped work landed, no schema-aware answers, and no production errors in the conversation when something breaks. Connect each one when its absence starts to hurt. See the Quickstart.

Which hosting, database and log tools can I connect?

Deployments: Vercel and Coolify. Databases: Supabase. Logs: Axiom. The setup for each is short and lives under project integrations. See Delivery and testing, Database and migrations and Log providers.

Does connecting them cost anything?

Connecting is free. What costs money is the background work that follows: reading commits and deploys, indexing, and Dev's searches. Each connection's guide names its own costs, and every charge lands on your organization's balance where you can see it. See Billing, usage and limits.

Can I disconnect one later, and what happens to what it learned?

Yes. Replace a token or disconnect the integration at any time from project settings, and DevSpec stops reading anything new. What's already recorded stays on the project: memories, linked deploys and history remain visible under their own retention. See Database and migrations, Delivery and testing and Log providers.

Decisions and approvals

What happens if new work conflicts with an earlier decision?

DevSpec shows which decision the work would undo and whose OK it needs. The agent can build it separately while it waits, but it must not merge into the project's tracking branch until it has that OK. You can give the answer on Review, on the affected record, or through Dev or your agent. See Review.

Can I try something without changing the team's decision?

Yes. Make the task an experiment and build it on its own branch. When you're ready, ask for it to land. Any decisions it would undo need their own OK before it becomes part of the project. See Tasks.

Do I have to leave the conversation to approve something?

No. Ask Dev or your agent to make the decision for you. They use your permissions. If another person's agreement is needed, that person gets a request to confirm it; someone else's agent cannot agree for them. See Review.

Memory

What is memory in DevSpec?

A memory is something your team has settled or learned that should outlast the conversation it came from: a decision, a convention, a lesson, a known risk. Once recorded, Dev and your agents can find it in every later session, so nobody re-explains why things are the way they are. See Memories.

How is that different from the memory in ChatGPT or Claude?

DevSpec memories belong to the project: the whole team and every connected agent can use the same record. They have versions and history, and decisions retain who asked for them. If new work would undo a decision, it waits for the right person's OK. See Memories and Review.

How do memories get made?

While the work is happening. When a session settles something worth keeping, Dev offers to record it and writes it when you agree. Your connected agent can record one when it learns something durable. Memories created during an automation wait in Review for a Maintainer to accept. Before anything new is saved, DevSpec checks for an existing memory that says the same thing, so the record doesn't fill with near copies. See Memories.

Can I edit or correct a memory?

Yes. Open it to change the title, summary or text, with earlier wording kept in history. Contributors can edit their own and other Contributors' knowledge directly; a change to a Maintainer's decision waits for approval. Removing someone else's decision needs their OK or that of someone who can decide in their place. See Memories.

What happens when a decision changes?

The memory is replaced, not silently edited. The old version becomes superseded and links to its replacement, which links back. Retracted memories stay in history but are left out of answers. Old decisions can't creep back into new work unnoticed. See Memories.

Does my agent actually use memories?

Yes. Dev and your agents receive relevant project knowledge and can search for more before answering or changing code. Current decisions guide new work; earlier versions remain available when you need to understand the history. See Memories.

Who can see and change memories?

Anyone on the project can read and search them. Reviewed Contributors propose knowledge, while Contributors and Maintainers can add it directly. Editing a decision depends on whose decision it is, and undoing it needs the appropriate OK. Viewers can discuss memories but cannot change them. See Memories and Organizations, members and roles.

What stops it remembering something wrong?

You can correct a memory while keeping its history. Checks raise descriptive knowledge that may be outdated, and changes that would undo a decision need the appropriate OK. Knowledge created by an automation waits as a proposal, and retracted memories stay out of current answers. See Memories and Review.

What do memories cost?

Browsing and opening memories is free. Searches and background checks are charged as Running your project. Asking Dev to write or explain one is part of Dev's reply. See Memories and Billing, usage and limits.