Export the weekly usage report as CSV.
waiting for a free slot (3 agents at most)
Notes & moderation
Screened for spam and personal data before an agent can read it.
Software factory by ABLD
Kanban-driven software delivery
Tokenalytics is a software factory for Claude and Codex agents. Drag an issue or a piece of user feedback into Todo, and an agent investigates, plans, builds, and opens a pull request. You approve the plan and decide what merges.
About ABLD
Tokenalytics is built by ABLD
ABLD (에이비엘디) is a Korean AI company founded in May 2026. We help teams put AI into practice through AI education, AX consulting, AI content systems, and AI software development.
Tokenalytics is ABLD's software product. Feedback and issues arrive as cards; Claude and Codex agents investigate, plan, implement, and open pull requests, and people review at each approval gate.
- Company
- ABLD (에이비엘디)
- Founded
- May 2026
- Founder & CEO
- Jeongmin (Tony) Lee
- Address
- 7F C143, 8-4 Hwangsaeul-ro 319beon-gil, Bundang-gu, Seongnam-si, Gyeonggi-do, South Korea
- Business registration
- 755-60-00875
- Mail-order business report
- 2026-Seongnam Bundang-B-0508
- Contact
- dev@abld.site
The problem
Agents write code fast. Delivery still needs a process.
One terminal per agent
Each agent session lives in one person's terminal, so work is hard to share, queue, or hand over.
No record of decisions
Why a change was made, who approved the plan, and which checks ran are scattered across chats and logs.
Limits stop work mid-task
Claude and Codex usage windows run out, and nobody sees it until an agent stalls.
Every comment becomes a card
Feedback and issues arrive in New. Drag a card into Todo to start an agent. Three agents run at most; the rest wait for a free slot.
Drag a card into Todo, or change its status, to start an agent.
New
4Reset time shows UTC instead of my local time.
waiting for a free slot (3 agents at most)
Notes & moderation
Screened for spam and personal data before an agent can read it.
Thanks for building this.
waiting for a free slot (3 agents at most)
Notes & moderation
Screened for spam and personal data before an agent can read it.
Could the menu-bar popover follow dark mode?
waiting for a free slot (3 agents at most)
Notes & moderation
Screened for spam and personal data before an agent can read it.
Nothing here
Todo
1In progress
2Codex quota window disappears after re-login.
waiting for a free slot (3 agents at most)
Notes & moderation
Screened for spam and personal data before an agent can read it.
Warn me before the five-hour window runs out.
waiting for a free slot (3 agents at most)
Notes & moderation
Screened for spam and personal data before an agent can read it.
Nothing here
Review
1Done
1Fully configurable, agent-powered
Define your lifecycle stages, then set rules for each stage and each type of work.
Issues, pull requests, and user feedback land on one board.
Open the Intake stage →Agents investigate each card, gather context, and say what is missing.
Open the Triage stage →The investigation becomes a plan you can read, edit, and approve.
Open the Planning stage →An agent implements the approved plan and runs your checks.
Open the Building stage →A separate agent reviews the pull request before you decide.
Open the Review stage →Verified work closes, and the source hears back.
Open the Done stage →The six-stage workflow, in practice
The rules every card follows, whichever agent does the work.
- 01
Investigate and plan
When a card enters Todo, an agent reads the codebase, gathers context, and writes an implementation plan.
- 02
People approve the plan
A person reviews and approves the plan. Nothing is built without that approval.
- 03
Build and open a pull request
The agent implements the approved plan on its own branch, runs the checks, and opens a pull request.
- 04
Iterate on review
When review feedback arrives, the agent revises the change and runs the checks again.
- 05
A gate at every major stage
Starting work, approving the plan, and accepting the review each wait for a person.
- 06
People decide the merge
The final merge is always a human decision on GitHub.
Team
Who builds Tokenalytics
Product
One board for the whole delivery loop
Factory board
A kanban board where each card carries its own agent session, files, and history.
Agents
Claude Code and Codex sessions that investigate, plan, build, and review in isolated worktrees.
Accounts & capacity
Provider limits and reset times decide when an agent can start, across all your accounts.
Observability
Tool calls, diffs, checks, and decisions, recorded on the card that caused them.
Solutions
Built for the work that piles up
Feedback to fix
User comments go to a feedback board. The good ones become pull requests.
Bug triage
Agents reproduce the bug before anyone writes a plan.
Maintenance backlog
Dependency upgrades, flaky tests, and cleanup, one approved plan at a time.
Solo builders & teams
One owner per board, with GitHub as the place teammates review.
Frequently asked questions
What is a software factory?
A software factory automates parts of the development lifecycle with agents. Tokenalytics uses staged automation: explicit gates at intake, planning, and review keep each change measurable and reversible.
Which agents does Tokenalytics run?
Claude Code and OpenAI Codex sessions, signed in with your own accounts. Model usage stays under your provider plans and their limits.
Can it merge code on its own?
No. Agents open pull requests. A person approves every plan and makes every merge decision.
Does it work with an existing codebase?
Yes. Connect an existing GitHub repository. Agents read the code, follow your checks, and work on separate branches.
Who makes Tokenalytics?
ABLD (에이비엘디), a Korean AI company founded in May 2026. About ABLD.
Is Tokenalytics available now?
Tokenalytics is in early access by invitation and is not publicly released yet.
Start building with Tokenalytics
Take work from feedback to a reviewed pull request, with agents you govern.