The problem
Context gets lost between chat sessions, terminal history, and task notes that live nowhere in particular. The bottleneck today isn’t code generation — it’s context control.
The solution
Tipatask is the project operating layer for AI coding agents: a task system, architecture memory, objective-refinement workflow, MCP interface, and local execution console that keeps AI-assisted work accountable to the project instead of scattered across chats, terminals, and disconnected boards.
The team plans and discusses the work in one place, and each member builds it on their own machine, using their agents. Everyone, agents included, works from one knowledge database, so context does not depend on what a single chat or terminal remembers.
Comments, mentions, and notifications on every task, with roles for everyone who follows the work.
A board, backlog, sprints, and epics that show what is planned, in progress, and delivered.
The desktop app next to their own copy of the code, with each task, its spec, and the project’s knowledge at hand.
Claude Code, Codex, or Pi, picked for each task and run in parallel, each reporting back what it changed.
How it works
Spec-Driven Development turns team input into a clear specification, an approved plan, and code you can review.
Executives, stakeholders, and developers share their plans, expectations, and desired outcomes.
Scaffold standalone Laravel site
Website style guide and insignia palette
Landing page L1–L15 sections
Describe them in your own words. The planning chat uses the project’s tasks and architecture to shape a clear, complete spec.
Build the public landing page: fifteen sections from the page map, in English and Ukrainian, styled with the site style guide. Every section must render in both locales.
Review the spec and proposed tasks. You approve what will be built before an agent starts.
1. @website/routes/web.php — serve the landing page at the root and the Ukrainian path from one localized route. 2. @website/public/css/marketing.css — copy the token block from the style guide.
1. @website/resources/views/landing.blade.php — build the fifteen page-map sections on their assigned grounds. Verify every section id renders in both locales. 2. @website/lang/en/landing.php — one key per visible string.
Files Addressed (3)
Agent edits during analysis (1)
Each task carries its approved specification, dependencies, and the project knowledge it needs.
1. @website/resources/views/landing.blade.php — build the fifteen page-map sections on their assigned grounds. Verify every section id renders in both locales. 2. @website/lang/en/landing.php — one key per visible string.
1. @website/lang/uk/landing.php — translate every key; identical values only for the untranslated-by-design set.
1. @website/resources/views/landing.blade.php — add a band between the personas and the roadmap with six speech-bubble tiles.
Launch an agent from the task to turn the approved spec into working code, on your machine, in your checkout.
[Task App] Auto-approved project MCP servers.
[Task App] Planning mode. Review the plan, then type /approve-plan and press Enter to continue.
/approve-plan
[Task App] Plan approved. Continuing in execution mode.
--- Process exited (code 0) ---
Review working code against the spec. The task keeps the result and updated project knowledge, so stakeholders can follow delivery.
pull_knowledge()
{ "pulledCount": 2 }
get_tag_architectures({ "tag_names": ["tt-website-laravel"] })
ai/architecture/tt-website-laravel.md
push_knowledge({ "file_key": "ai/architecture/GENERAL.md" })
{ "pushed": 1 }
KB synced — 1 pushed, 2 pulled
Task management and communication
Give executives, stakeholders, and developers one shared view of plans, progress, and results. Plan, assign, and discuss work on the same task record.
Project memory
Architecture knowledge lives with the project, so every agent starts informed and leaves the project better documented.
Swappable parallel agents
Spec-Driven Development turns your approved spec into working code. Each agent works from the same clear scope and project knowledge.
What makes us different
Both contrasts reduce to one discipline: Spec-Driven Development. A human approves the specification before execution, and the result is reviewed against it.
Prompt-to-app tools optimize for something that runs quickly, with no durable plan behind it. Tipatask governs ongoing work, task by task, with a record attached to each one.
Most task tools were built for humans coordinating with humans, with AI added as a chat feature. Tipatask starts from a task record an agent can read and act on directly.
Who it’s for
Tipatask remembers the project across sessions — the plan, what got tried, the architecture — without a workflow heavier than the work itself.
Task specs an agent can execute, and agent work that is reviewable the same way human work already is.
One surface for objectives, architecture context, and execution.
News
Version 0.9.21 Changelog
Tipatask 0.9.21 gives AI agents a new way to confirm their own work. On macOS, task runs powered by Claude, Codex and Pi can now capture screenshots and click or type inside applications. This lets them check that interface changes actually behave as intended. In Plan mode, agents can only…
Version 0.9.19 Changelog
Tipatask 0.9.19 makes whole projects portable. A new export option packs a project's knowledge base and all open tasks, with their sprints, tags and statuses, into a single zip archive. Importing it creates a fresh project through a short wizard that asks for a name, folder and agents. Files…
Version 0.9.17 Changelog
Agent permission modes. Image-generation tasks with OpenRouter. Objectives keep their source material.
Roadmap
Every shipped feature, with the tasks that built it, is on the roadmap page.
94% of Tipatask has been built using Tipatask
94% built with Tipatask 6% by hand
Tasks by Agent
Desktop app
Tipatask runs on macOS, Windows and Linux. Pick your system and download the installer straight from the latest release.
It signs in with the same account as the web app. New here? Create one first.
Create your accountStill in beta
TipATask is still in beta, so expect rough edges and features that change between releases. Updates are not automatic yet: check back here for new versions.
First launch
The Windows installer is not code-signed yet, so SmartScreen may show “Windows protected your PC”. Choose More info, then Run anyway.
First launch
The .deb and pacman packages install like any other package. For the AppImage, allow it to run as a program in the file’s Properties → Permissions, or run chmod +x on it. If it does not start, install the libfuse2 package (libfuse2t64 on Ubuntu 24.04 and newer).
Testimonials
Nothing of that would be required, you know.
Easy peasy!
Team work is the key.
Delegation is all about trust.
No more tasks hanging.
I love the big-picture view.
