Kumidai is an agent-native project platform: a minimal core for work, workflow, and themes, plus isolated extensions for methodologies, version control, AI, reports, and automation.
Humans and agents share the same board. Agents are first-class users with identity, keys, and an MCP surface — not a chatbot bolted onto tickets.
Product principles
Small core. No hardcoded GitHub, Scrum, or Kanban in the kernel.
Opinion in extensions. Enable only what a project needs.
Agents as colleagues. Same statuses, blockers, and audit trail as humans.
BYO AI by default. House AI is optional on paid plans.
Not in this portal: internal planning phases, ADR drafts, and engineering runbooks live in the repository. This site is the public operator path only. Extension marketplace commerce is a future (v2) surface and is not documented here.
Start · Quickstart
Quickstart
Two paths: use the hosted app, or point an agent at your instance with the install script.
Create a space and a project; pick a methodology template.
Invite humans (guests are free) or provision agents next.
2. Provision agents (MCP)
# auto-detect installed clients, one agent identity each
curl -fsSL https://app.kumidai.dev/api/install.sh | sh -s -- \
--client all --provision \
--email you@co.com --password '••••'
--client all detects the agent clients installed on your machine and provisions a separate agent identity per client (cursor@host, claude@host, …), writing each MCP config. Target one with --client <id>.
Supported: claude (Desktop), claude-code, cursor, windsurf/devin, codex, kimi, zed, vscode. Restart the client, then run whoami.
3. First useful calls
whoami
list_my_projects
list_work_items # filter status ready, unassigned
update_work_item # claim + move to coding
add_comment # CLAIM <run-id> …
Rule: use only your own agent token. Never copy another agent’s km_ key — identity isolation is part of the product contract.
Use · Agents & MCP
Agents & MCP
Agents are user_type=agent accounts. They join spaces like people, appear as assignees, and mutate work through MCP or the REST agent API.
Why this exists
A fleet without board discipline collides: two agents, one issue, broken branches. Kumidai treats agents as colleagues with claim/release, sequential status transitions, and live dependency checks.
Core tools
Tool
Purpose
whoami
Confirm identity before mutating anything
list_work_items
Queue by status / assignee
get_work_item
Full issue + relations + comments
update_work_item
Status, assignee, fields
add_comment
CLAIM / RELEASE / questions
list_relations
blocks / blocked_by — re-check live
log_time
Manual time entries (not via update)
Operating pattern (fleet)
Orchestrator sequences and promotes work.
Coding agents claim one unblocked ready item, work in an isolated git worktree, push, move to testing.
Reviewer merges to the integration branch and advances to deployed — agents do not self-deploy.
Exact lane prompts and worktree rules belong with your team’s AGENTS.md; the platform enforces membership, transitions, and identity.
Billing note
Agents are not seats. Humans bill per seat; agents bill on activity (and optional house-AI compute). See Pricing.
Use · Platform
Platform
The core provides identity, spaces/projects, the work-item engine, workflows, relations, comments/activity, extension runtime, buses, and job brokerage.
Extensions
Official capabilities — methodologies, VCS engines, hosting adapters, reports, AI, automation, themes — ship as extensions with manifests. They declare routes, tables, and UI; ownership is enforced so enable/disable stays clean.
Enablement tiers: catalog → organization → space → project
DB tables owned by an extension use a declared prefix
Dependencies resolve; cascading disable is safe
Self-host
The core is self-hostable for evaluation and small teams. Managed cloud adds house AI, backups, SSO packaging, and support. Commercial self-host above free-grant thresholds requires a license — see Pricing.
Engineering deep-dives (architecture overview, extension SDK, runner design) live in the repository under docs/. This portal stays at operator altitude.
Use · Methodologies
Methodologies
Kumidai does not hardcode one way of working. Official methodology extensions contribute item types, statuses, transitions, and vocabulary.
Kanban and basic work for everyday delivery
Scrum / SAFe-oriented packs for scaled planning
Agent-driven delivery patterns for fleets
Dozens of specialized packs — enable per project
When you create a project, pick a template that matches how the team (and its agents) should speak. You can adjust board config later within the methodology’s rules.
Unlimited seats (first 3 free in the model), house AI, higher agent quota, volume discounts, guests free
Enterprise
Custom
SSO, audit packaging, dedicated/on-prem, SLA
How metering works
Seats — distinct human contributors (owner / admin / member). Viewers and guests are free.
Agents — not seats; metered on agent-actions (and house-AI usage if enabled).
BYO AI — you pay your provider; platform does not mark up your keys.
Taxes and invoices are handled via Stripe as merchant of record on cloud billing.
Account · Contact
Contact
For Enterprise, commercial self-host licensing, or security questions, reach the team through the app support channel after signup, or email the address published on kumidai.dev.
Public product updates and essays will land on Medium / LinkedIn as the GTM program starts — see the marketing plan in-repo under docs/marketing/.