What is DevSpec

For why we built DevSpec and the problems it solves for teams building with AI, read Why DevSpec; if you want to start setting up right away, go to the Quickstart.

DevSpec is where your team, your knowledge and your agents build software together, made for AI-native development from the ground up. It connects two things: DevSpec in your browser (on desktop and phone), where your team talks work through and keeps a shared record of the project, and your coding agents (Pi, Claude Code or Antigravity CLI), which run on your own computers and write the code.

Here is how the pieces fit together from start to finish:

1. You talk a change through in a session with your teammates, Dev (DevSpec's assistant in every session), and any agents you bring into the room. Dev and your agents already know your codebase, database schema, production logs and past decisions, so you can figure out what to build in plain words.
2. Agreed work becomes tasks and memories. When the room settles on a plan, Dev turns it into tasks with clear acceptance criteria (grouped under a brief when several items belong together) and records any durable decisions or conventions in the project's shared memory.
3. Your agent builds the change on its own branch. When you hand a task to an agent, it reads the task, the acceptance criteria and the conversation behind it, writes and tests the code on a separate branch, and stops at Implemented with concrete evidence for each criterion. If the change would undo a teammate's earlier decision, DevSpec catches the conflict and keeps the work on its branch until the right person gives their OK.
4. DevSpec follows the work through to deploy and Done. When your host deploys the branch (or when someone pushes a commit directly to GitHub), DevSpec links it back to the tasks and goals it belongs to, and marks the task Done once its acceptance criteria are verified.

The rest of this page walks through each part of DevSpec in plain English and links to the full guide for each one.

What DevSpec does, and what your agent does

DevSpec and your coding agent split the work cleanly so a person is always in charge of what gets built. DevSpec plans, tracks and remembers: it indexes your code, holds the sessions where decisions get made, turns agreed work into tasks, protects settled decisions, and watches commits and deploys so it knows what shipped. Your agent writes the code: when you ask it to take a task, it reads everything DevSpec knows about the work, makes the change on its own branch in your repository, and records what it did and how it tested it. Neither DevSpec nor Dev ever edits your code, commits or pushes on its own.

What is a project?

A project is the workspace for one product, service or initiative, holding its sessions, tasks, goals, memories, documents, connected integrations and team members in one place. A project can track several GitHub repositories at once (such as a web app, a backend and shared packages) and search across all of them together, or start with no repository at all while you plan from documents and conversations. Each project has its own members, roles and rules for how Dev and agents behave. See Projects and Project Overview.

Sessions and Dev

A session is a shared conversation room inside a project where you, your teammates, Dev and your connected agents can talk by text or voice. Dev is the assistant in every session: it searches your code, tasks, memories, past sessions, database schema and logs, and gives you an honest technical or product view rather than just agreeing with you. Inside a session, you can also run team polls, track multi-step plans, send directed questions to a specific teammate when a call needs their answer, and keep a running Still to Discuss list so open questions never get lost in scrollback. See Sessions, Dev and Voice.

Tasks and verification

A task is one tracked piece of work (a Feature, Improvement, Bug, Investigation or General item) with an intent explaining why it matters, a description of what to do, and acceptance criteria stating how you will know it is finished, optionally grouped with related tasks under a brief. Tasks move from Open to In progress while an agent works on them, and stop at Implemented when the agent finishes its branch. An agent never marks its own work Done: instead, it records concrete evidence for each criterion (such as an API test, CLI test, schema check or browser check) and flags anything that needs a person to confirm, so tasks only reach Done when the proof is in or someone verifies them. See Tasks and How work flows through DevSpec.

Agents and automations

When you start Pi, Claude Code or Antigravity CLI (agy) on your computer in a project folder, it connects to DevSpec and appears on the project's Agents page. You can give it instructions in your terminal, bring it into a session from your browser or phone, or hand it a batch of tasks (including picking a safe parallel set that will not collide on the same files), and if you choose to share an idle agent with teammates, they can ask it to build or investigate work too. For recurring chores, Automations let Maintainers write plain-English instructions that run on a schedule or when a deploy fails, using a connected agent with the permissions you choose. See Connect your agent, Working with your agents and Automations.

Memories and documents

A memory is a piece of knowledge your team has settled or learned that should outlast the conversation it came from: a decision, a convention, an insight or a risk. Memories are proposed during sessions or recorded by agents as you work, and DevSpec automatically supplies relevant memories to future sessions and agents so nobody has to re-explain how the project works or why a choice was made. Alongside memories, you can upload documents (like PDFs, specs and screenshots) for Dev and agents to search and cite, and ask Dev or an agent to write longer artifacts such as plans, guides, runbooks and decision records. See Memories and Documents and artifacts.

Protecting your team's decisions

Because everyone on a team and their agents share one codebase, DevSpec continuously checks new work and edits against your project's existing tasks, decisions and conventions. If a task or code change would undo a settled decision, DevSpec raises a conflict and shows whose OK is needed, while still letting the agent build the work safely on its own branch (or as an explicit experiment) until that person or the project lead decides. The Review page collects these conflicts alongside duplicate items, proposals from Reviewed Contributors or automations, and memories that look out of date after code changes, and you can resolve any of them either on the page or simply by telling Dev or your agent in conversation. See Review and Rules and instructions.

How we use your repository

When you connect a GitHub repository, DevSpec clones your chosen tracking branch and keeps a search index of the code up to date as commits land, so Dev and your agents can find the exact files and symbols they need. DevSpec also reads every commit pushed to your repository: commits tagged with a task ID link straight to that task, while unlinked commits that land without a task are inspected automatically so quick fixes get written up as tracked work and checked against your team's decisions instead of sneaking through unnoticed. See Repositories and GitHub, Code indexing and Commits.

How we use your database

Connecting your Supabase database makes Dev and your agents schema-aware, so the SQL, migrations and backend code they write match your real tables, columns and policies instead of guessing. DevSpec reads the structure of your public schema (tables, columns, types, relationships, indexes, functions, triggers and row-level security policies), your applied migration history, and Supabase's security and performance advisors. It never reads your application rows and never writes to your database. See Database and migrations.

Why connect logs?

Connecting your logs lets Dev see what your application actually does in production, not just what the source code looks like on disk. With Axiom connected, Dev can search the datasets you allow, look up the real errors behind a bug report, trace a failure back to the code that caused it, and check logs after a deploy to confirm a fix worked. Logs are optional and you choose exactly which datasets Dev may search. See Log providers.

Delivery, goals and analytics

When you connect Vercel or Coolify, the Delivery page links each deploy back to the commits and tasks inside it, lets you know when shipped work is live and ready to check, and can notify your team or trigger an automation if a deploy fails. To help leaders and teams see the bigger picture without status meetings, Goals track high-level milestones across tasks, Analytics shows project throughput and spend, and the Changelog keeps a clean feed of shipped and verified work. See Delivery, Goals, Analytics and Changelog.

Who can do what

Every person in your organization has an organization role (Owner, Admin, Member or Guest) and a role on each project they belong to:

• Viewer: can read the project and take part in sessions, without changing tasks or memories.
• Reviewed Contributor: can propose tasks and knowledge for a Maintainer to accept in Review, which works well for clients or new collaborators.
• Contributor: can create tasks and memories directly; changes that would undo agreed work or someone else's decision wait for the right person's OK.
• Maintainer: manages project settings, integrations and automations, and can accept proposals.

Each project also has a Project Lead (a Maintainer) who can give the final OK on any decision conflict across the project. See Organizations, members and roles.

What it costs

Each organization has one prepaid balance in dollars, and your first organization starts with $15 on us so you can try everything before adding funds. There are no per-seat fees and no monthly subscription tiers: DevSpec only charges your balance for the work DevSpec itself performs (such as Dev's replies, voice, code indexing and background checks), while your coding agent's model usage is billed directly by your own AI provider. Owners and Admins can set monthly spending limits per person and per project at any time. See Billing, usage and limits.

What DevSpec works with today

DevSpec runs in your browser on desktop and phone, with no separate mobile app to install.