agentic ide // open source // fork of orca

Your IDE just hired a team

Wakii is an agentic IDE with a built-in superpowers team — plugin bundled, kit auto-installed to ~/.claude, enabled by default. Describe an idea and the pipeline runs: plan, parallel agents, verify gates, one clean PR.

9
agents, role-gated
0
setup steps
1
PR per story
wakii — zsh
wakii ./my-project
› resolving project context … done
› bundled plugin loaded · kit → ~/.claude · team enabled
› 9 agents online · superpowers panel ready
› memory restored: 47 verdicts, 12 learned patterns
/superpowers "ship the next feature while I get coffee"
features // live demos

Watch the team work

Every cell below is a live piece of the Superpowers panel, mid-flight on a real story. Hover to pick one up — this is what "agentic IDE" actually looks like.

bracket canvas FI-289 · wakii-site 2 running 1 done
epic
spec & plan
parallel SFs
FI-289 epic
agentic IDE landing
coordinator
SF-1 done
spec + design direction
designer · spec-critic
SF-2 running
bento landing EN+VI
task-executor
SF-3 queued
docs × 2 locales
SF-4 running
perf budget ≥ 90
verifier
SF.002 — bracket canvas: epic → SF nodes with dependency edges, drawn live as the pipeline resolves them.
superpowers team4 working
task-executor
ships SF slices
working
code-reviewer
reviews at B3
reviewing
verifier
tests + browser
47 green
spec-critic
attacks specs
online
plan-critic
stress-tests plans
online
phase0-impact-analyst
blast radius
online
security-audit
leaks & injection
online
rollback-fixer
clean reverts
online
designer
direction + UI
drafting
SF.003 — 9-agent team: role-gated. No agent ships outside its job.
story ops — gatesB3 in progress
B0
browser test
B1
code + tests
B2
plan
B3
B3
review
B4
B4
merge
B5
B5
done
Story Ops: six gates per story. B3 review — APPROVE · 47/47 tests.
wakii memory — verdicts.logsession 47
09:41:07 B3 SF-1 PASS — matches design tokens
10:02:33 B3 SF-2 APPROVED — "extracted cleanly"
10:14:58 watchdog stall → resumed on SF-3
10:15:02 pattern learned: verify-first · 12 stories
SF.005 — memory loop: verdicts persist. Story #20 never re-argues story #3.
watchdog — livemonitoring
▸ watching SF-3 hand-off … silence threshold 45s
⚠ stall detected — executor idle 52s, checkpoint found
▸ resuming SF-3 from last checkpoint …
✓ SF-3 back on track — no human involved
SF.004 — watchdog auto-complete: stalled work finishes itself. You get coffee.
figma → verify pipelinedesign → shipped, diffed
capture frame + tokens captured, committed to the repo
implement tokens-only, component map before code
verify screenshot diff vs capture, side by side
visual diff: 0 unexpected deltas
SF.006 — figma-to-verify: the build is checked against the design, pixel by pixel, before merge.
bootstrap 0.0s

Bundled plugin

Superpowers ships inside the app. No marketplace, no version juggling.

install ~/.claude

Kit auto-installs

Skills land in ~/.claude on first launch. You never touch a config file.

config enabled=true

Enabled by default

The team is on when you open Wakii. Opt out per-project — not opt in.

pipeline

cat workflow.md

One pipeline from spoken idea to merged PR — visible end to end in the Superpowers panel.

idea impact plan (linear subtasks) parallel SFs verify gates 1 PR / story
▮ highlighted stage = agents executing concurrently
who does what

Nine role-gated agents run every story. Analysts frame the problem, critics attack the spec and the plan, an executor builds in an isolated worktree, a designer prototypes the UI before code — and an independent reviewer plus a verifier check everything the executor claims. No agent approves its own work.

six gates per story

Every sub-feature passes the Story Ops gates before it counts. Gates are tool-enforced — a gate that could self-approve would not be a gate.

B0
browser test

A real three-tier walkthrough — DOM, screenshots, click-through flow — before anything is called done.

B1
code + tests

Implementation complete, tests green on the touched surface. No green, no gate.

B2
plan

Every plan step ticked off with evidence — nothing silently skipped.

B3
review

An independent reviewer attacks the diff: bugs, security, scope creep.

B4
merge

Merge into the story destination branch, audit trail recorded.

B5
done

The issue closes only after verification is clean — tracking truth matches the repo.

The pipeline runs in the open. Track every slice, gate and verdict live in the Superpowers panel and the bracket canvas — while the watchdog makes sure nothing stalls.
philosophy

Eight pillars, one pipeline

The mechanics are public — the reasons they are built this way matter more. Condensed from the full story workflow; each pillar has a deeper write-up in the docs.

01
Analyze once, inherit many

Deep analysis is expensive, so it runs exactly once at the epic level — every sub-feature inherits the results through a context pack.

02
A team with separated powers

The PM specs but never codes; the developer builds but never approves its own work; the tester hunts failures but never fixes them.

03
Gates, not trust

Tool-enforced gates plus a real browser walkthrough demand evidence — "the agent says it works" is not a verdict.

04
Humans own the irreversibles

Architecture decisions and the merge to your branch are human gates — nothing irreversible happens without a person saying yes.

05
Tiers and one destination branch

Dependencies are explicit tiers; every tier merges into one story branch, so parallelism never means divergent histories.

06
Watchdog: idle is not dead

A silent agent might be running a long build — the watchdog checks three layers before resuming a stall from its last good state.

07
Memory and the learning loop

Every task ends with a ritual — what broke, what fixed it — so the next story starts smarter instead of rediscovering the same trap.

08
Defensive by design

The workflow assumes it will be the one making the mistake — so mistakes are cheap, visible, and revertable by default.

read the full philosophy

download

get wakii

Installers for macOS and Windows, or build from source — and a mobile companion for your agent team.

download wakii.desktop primary
Wakii-1.4.199-arm64.dmg · orca-windows-setup.exe
Unsigned build — macOS apps outside the App Store need to be allowed in System Settings → Privacy & Security.
Unsigned build — SmartScreen may show a warning. Choose "Run anyway".
build from source dev

Wakii is a desktop app built from an open repo. Three commands from clone to the Superpowers panel — no configuration along the way.

read the guide
wakii in your pocket ios · android

A companion app for iOS and Android — check on your agents, approve gates, and send tasks from anywhere.

about the mobile app

Prefer an installer? Grab Wakii for macOS or Windows straight from the download page.

faq

frequently asked

Is Wakii really zero-setup?
Yes. The plugin is bundled in the app and the skills kit installs itself to ~/.claude on first launch, enabled by default. Open Wakii — the team is already there.
How is this different from an AI plugin in my editor?
A plugin gives you a chat window. Wakii gives you a team: nine agents with defined roles, gates between them, a watchdog that completes stalled work, and memory that compounds across sessions.
What does "one PR per story" mean?
Each story — however many agents and spec slices it takes — lands as one clean, reviewable pull request with verification behind it. No PR storms, no half-merged states.
What is Wakii's relationship to Orca?
Wakii is an open-source fork of Orca, which is MIT licensed. We credit the upstream project fully and build the superpowers team on top of it.

only the short version — the full FAQ