Changelog
This page lists what is new in tavya, newest first: the features, providers, plans, commands and docs added each month.
November 2026
- Open source — tavya's source code is public under the GNU AGPL-3.0, and you can host it yourself on your own server.
- Free trial — your first organization gets $5 of free compute for 14 days, so your first tasks need no card.
- Included model — tasks start on a model tavya provides, paid from your compute, with no agent login; any OpenRouter model works too.
- Repositories without GitHub — tavya can host a project's Git repositories itself; clone them over HTTPS, and move one to GitHub later.
- Autonomous projects — switch a project to fully autonomous, and agents plan the work, answer questions, review results and decide requests within a budget.
- Files — a project's Files tab shows its repositories, wiki and data as one folder tree; share a file or folder by public link.
- Answer from the list — approve a task's requests and answer its quick questions from its row in any task list, without opening it.
- More ready-made tasks — the task store adds weekly dependency updates, daily task tagging and weekly upkeep of the project wiki.
- Claude Fable 5.2 — choose Anthropic's Fable 5.2 as the model of a task that runs Claude Code.
- Team seats — add or remove seats on the Team plan in Plan & billing, and keep a plan you had set to cancel.
- Free plan computers — on the Free plan, tavya's own computers go up to 2 CPUs, 4 GB of memory and 40 GB of disk.
- Simpler console — redesigned forms, a sidebar without a top bar, bigger targets, and decisions named by what they allow.
- Faster review files — the images, videos and other files an agent attaches for review open faster.
- New guides — the docs add guides on parallel agents, scheduled tasks, agent review, agent payments, ignored files and self-hosting.
October 2026
- tavya CLI — put a project or a task on your own machine with its data and secrets, and start, follow and confirm tasks from the terminal.
- tavya's own computers — run tasks on computers tavya provides, charged to your plan's compute, with no provider account of your own.
- Four more cloud providers — connect your own Fly.io, Hetzner, AWS or Google Cloud account, for larger disks, lower prices or your own region.
- Any provider — leave a task's provider on Any, and tavya uses the first computer that fits and has room.
- A computer per task — choose each task's CPU, memory and disk, grow them while it runs, and move a running task to another provider.
- MCP and tokens — connect Claude Code, Codex, Cursor or another agent to tavya over MCP, and create personal tokens for scripts and CI.
- Chat bots — ask a project for work from Telegram, Slack, Discord or WhatsApp; each request becomes a task, answered in the same thread.
- Events and webhooks — start tasks from GitHub activity, incoming webhooks, or events that another task emits.
- Task templates — save any task as a template, and start new tasks from it in the console or from the CLI.
- Task store — browse ready-made tasks, read what each one does, then run or install it in a project.
- Several agents, one conversation — type @ in a task to call in another agent or a person; they share one conversation and take turns.
- Several pull requests per task — one task can work on several branches and open a pull request for each of them.
- Fork contributions — work on a repository you cannot push to: the agent works on your fork, and the pull request goes upstream.
- OpenCode in the cloud — OpenCode now runs on a task's cloud computer, as Claude Code and Codex do.
September 2026
- Connected apps — connect apps such as GitHub, Slack or Google through MCP or Composio, and choose which ones each task's agent may use.
- Cards and a budget per task — give each task its cards and one total budget; a payment above the budget waits for your approval.
- Insights — see what your organization shipped, how long agents worked, and the tokens each model used.
- Share a conversation — share a task's conversation by public link, when your organization and project allow it.
- New models — Claude Opus 5.5 and OpenAI's GPT-6 models are available for tasks that run Claude Code or Codex.
- One Tools picker — choose a task's MCP servers and connected apps in one searchable Tools field in the task form.
- Upload data — add single files or a whole folder to a project's data from your browser.
- Delete your account — request deletion of your account from your profile; tavya confirms what is removed and what is kept.
- Fullscreen conversation — read an agent's conversation in fullscreen, and press Escape to leave it.
- A new name — the product is now called tavya, and its home page follows your light or dark theme.
August 2026
- Plans — hosted plans set how many agents run at once and how much storage you get; manage yours in Plan & billing.
- Setup guide — a new organization gets a guide through connecting accounts, creating a project and running a first task.
- GitHub App — connect GitHub by installing tavya's GitHub App on the repositories your agents may work on.
- Responder — choose who answers an agent's questions while it works: you, another person, or an agent.
- Projects without a repository — create a project with no repository, for work that is not code.
- Your own storage — keep a project's data in tavya's storage, or in an S3 bucket of your own.
- Work locally — continue a task on your own machine, with its branch and its agent's conversation.
- Bring a conversation — start a task's agent from a Claude Code or Codex conversation that you upload.
- New data from a task — when a task produces files worth keeping, adopt them as project data at review.
- Export your data — download a JSON archive of your profile, or of a whole organization, from settings.
- Password store sync — connect a pass password store kept in Git, and the vault stays in sync with it.
- Wiki links in tasks — type [[ in a task to point its agent at a page of the wiki.
- Project folders — group an organization's projects into folders, and folders inside folders, in the sidebar.
July 2026
- Cloud computers — each task runs on its own cloud computer, on E2B or Daytona, and pauses while it waits for you.
- Sub-tasks — a task can split its work into sub-tasks, each with its own agent, and wait for them to finish.
- Agent review — let an agent review a task's result, alone or as one of several review steps.
- Several attempts — run the same task several times at once, and keep the attempt you prefer.
- Several repositories — a project can hold several repositories, and one task can change them together.
- GitHub pull requests — a task's work goes to GitHub as a pull request, and merges when you confirm it.
- Vault — store logins, API keys and passkeys, choose which ones each task may use, and approve an agent's requests for more.
- Cards for agents — save a virtual card of your own, with a limit, for agents to pay with.
- Wiki — every project and organization has a wiki of skills, memories and prompts that agents read and update.
- OpenCode — run tasks with OpenCode, as well as with Claude Code and Codex.
- Images in prompts — paste or drag images into a task's prompt for the agent to see.
- Task numbers and links — every task has a number in its project, and an address you can bookmark.
- Keyboard navigation — move around the console and act on tasks without leaving the keyboard.
- Search — find tasks by what was said in their conversations, and save a search as a view.