Meet your team
Start with the outcome you want and the roles needed to achieve it. Hire a developer,
reviewer, QA engineer, or writer, and give your Supervisor an objective. The Supervisor
delegates tasks, coordinates handoffs, and brings the team's results back together.
Workers can use different supported AI providers in the same project. Automatic agents follow
model policies: start with Plenipo's suggested role preferences, then set models and effort for
your organization, department, role, or agent. Unset effort uses the model or AI tool's default.
You control permissions and approvals, and can inspect why a model was chosen.
For example, hire a developer to build a services page and a reviewer to check it. Your policies
could assign Claude Code to the developer and Codex to the reviewer. Separate working copies,
owner terminals, read-only code Watch, test results, and task history support the work.
Real app capture from a development acceptance run, using synthetic test data. The latest
release may look a little different. Screenshot provenance.
Start here
Windows 11 x64. Open the latest release,
download its Plenipo_<version>_x64-setup.exe asset, and run the per-user installer.
Read that release's notes for changes, known limits, and installation details.
- Install and sign in to at least one supported AI tool.
- In Plenipo, open AI tools and choose Re-check. Confirm the tool is Ready.
- Create a department and project in Organization, add roles, and review their permissions.
- Give a Supervisor a small objective, then follow its work and review approval requests.
Try: “Review this project's README and report unclear setup steps. Do not change files.”
Use a role with read-only permissions for that first run.
Release status:
The badge always shows the latest published installer. The main branch can be ahead of it: a
merged version bump is not a published release.
What you can do
| Your goal | How Plenipo helps |
|---|
| Delegate a development task | A Supervisor coordinates roles such as Developer, Code Reviewer, QA Engineer, and Documentation Writer. |
| Choose the right AI tool for each job | Set model preferences and effort by role, with explained model selection and review-provider policies. |
| Keep access deliberate | Give roles permissions to read, change, run, or use git. Supported actions go through Guard and approval rules. |
| Work on websites | Use Plenipo's separate Edge or Chrome profile, allowed sites, approval cards, and owner takeover. |
| Inspect your servers | Configure Linux/Unix SSH servers, pin their identities, and decide which commands roles may use. Production commands ask every time. |
| Understand the result | Review local task history, model choices, approvals, changes, tests, and findings in the Ledger and Activity trail. |
Browser, desktop, and SSH controls include Stop all and takeover/disconnect controls.
SSH starts off. Sending, buying, and signing in ask by default; Settings switches can allow them
without asking on allowed websites. Read permissions and known limits
before granting access.
Your tools, your sign-ins
| AI tool | Current integration |
|---|
| Claude Code | Native Windows tool, signed in with your Claude account |
| Codex | Native tool from the Codex package, signed in with ChatGPT |
| Grok | Grok Build, signed in with its supported subscription |
| Kimi | Kimi Code, signed in with its subscription; file access goes through Plenipo |
| Ollama | Signed-in cloud models, text conversations only; no file or command tools |
| Antigravity | Google's Antigravity CLI, signed in with Google; text conversations only |
| GitHub Copilot | GitHub's Copilot CLI, signed in with GitHub; paid extra use must be off; text only |
Plenipo uses account sign-ins for these tools and rejects API-key authentication for them. Provider plans, usage limits,
and availability still apply.
Paid per use, only if you choose it: with Let workers use paid AI keys switched on and a
monthly spending cap set, you can add your own key for OpenRouter, or for an AI company's own service
(Anthropic, OpenAI, xAI, Moonshot AI, Google, DeepSeek, Z.ai, MiniMax, Mistral, and Alibaba Cloud).
The key is typed only into Plenipo and kept in Windows Credential Manager, a paid model runs only
where you list it, and Plenipo never starts a paid task that could pass a cap
(paid keys and spending caps). Installation and adapter details.
Local-first means local control and records. AI requests and task context still go to the
connected providers. This is not an offline-inference app. Data and privacy FAQ.
How the organization works
flowchart LR
P["President: you"] --> VP["VP"]
VP --> M["Manager"]
M --> S["Supervisor"]
S --> D["Developer"]
S --> R["Reviewer"]
S --> Q["QA and docs"]
You set the outcome. Leads pass work down to a project team. Workers act within their role's
permissions and supported tools, and hand back a result you can review. Development objectives
use their own working copies and branches. Titles can be personalized without changing roles.
Free today; editions planned
The current release has no license check or edition limits. The proposed Free/Pro split,
including subscription pricing and business departments, is a plan for a future release.
See Free and Pro for the full proposal. AI provider subscriptions are separate.
From v1.9.0, every copy (Free and Pro alike) checks Plenipo's GitHub Releases once a day for a
new version and tells you when one is ready. The check sends nothing about you or your work, and
nothing installs until you choose Install now (ADR-038, updates).
Plenipo is source-available under the Elastic License 2.0. Read the license for
its permissions and restrictions and CONTRIBUTING.md
for contribution terms.
Build from source
Install the Windows prerequisites first, including
C++ Build Tools, the pinned Rust toolchain, Node.js, pnpm, and WebView2.
git clone https://github.com/Seckcey/plenipo.git
cd plenipo
corepack enable
pnpm install --frozen-lockfile
pnpm dev
No API keys, provider login, or .env file is needed to build and open the app.
Running workers requires a supported signed-in tool. Windows 11 x64 is the product target;
Linux is used for development/tests, and macOS is not a current target.
Help shape Plenipo
Useful first contributions are a reproducible Windows bug report, a clearer setup step, or a
sanitized real-app screenshot. Browse open issues
and discuss larger changes before starting. CONTRIBUTING.md explains the checks,
project conventions, and licensing terms.
- Get help or ask the community.
- Report a bug.
- Read the roadmap, including what is published, merged, planned, or proposed.
- Security concern: read SECURITY.md first. Do not publish sensitive details in an issue or discussion.
- Star the repository if you want to find it again; use Watch → Custom → Releases for release notifications.
Documentation
Read the terms of service and
privacy statement. The same statements are available on the
website: Terms ·
Privacy. For private privacy or legal requests, email
[email protected].
Stack
| Layer | Technology |
|---|
| Desktop shell | Tauri 2 |
| UI | React 19 + TypeScript + Vite |
| Privileged core | Rust (stable) |
| Package managers | pnpm (JS), Cargo (Rust) |
| Primary target | Windows 11 (NSIS installer) |
Common commands
| Command | What it does |
|---|
pnpm dev | Run the desktop app in development mode |
pnpm build | Build the release app and Windows installer |
pnpm check | Versions, format, lint, typecheck, frontend tests |
pnpm test | Frontend unit tests (Vitest) |
pnpm typecheck | TypeScript typecheck for all packages |
pnpm lint | ESLint, the color and page-policy checks, and doc links |
pnpm format | Prettier (write) |
pnpm bindings | Regenerate TypeScript DTOs from Rust (packages/types) |
pnpm e2e | End-to-end tests against the release build (see setup guide) |
cargo test --workspace | Rust unit tests |
cargo clippy --workspace --all-targets -- -D warnings | Rust lint |
cargo fmt --all | Rust format |
Repository layout
apps/desktop/ React + TypeScript UI (Vite)
apps/desktop/src-tauri/ Tauri 2 Rust backend: typed command boundary, capabilities
crates/core/ Plenipo Core: provider-neutral domain types and shared DTOs
crates/ledger/ Plenipo Ledger: SQLite system of record, migrations, event trail
crates/liaison/ Plenipo Liaison: handoff protocol, context packets, replies between
workers
crates/runtime/ Plenipo Runtime: process supervisor, launch profiles, policy,
agent runtime adapters (Claude Code, Codex, Grok and Kimi over ACP,
Ollama, Antigravity, GitHub Copilot) and sessions
crates/workforce/ Plenipo Workforce: organization engine (positions, teams, oversight,
role templates), live snapshot, role routing for Liaison
crates/router/ Plenipo Router: model registry, role model policies, explained
choice of AI tool and model, usage limits
crates/guard/ Plenipo Guard: permission registry and sets, policy engine, folder
confinement, command rules, sensitive actions, secret redaction,
servers and the kinds of commands on them
crates/capabilities/ Capability broker: grants, Plenipo's tool server and relay, file,
program, git, and GitHub tools, working copies (a branch per
objective), approvals, Vault (OS credential store), Plenipo's
browser, screenshots, the screen, mouse, and keyboard, SSH to the
owner's servers, and the control center (sign, Stop, Take over,
Disconnect)
packages/types/ TypeScript DTOs generated from Rust (do not hand-edit)
tests/e2e/ End-to-end tests driving the real app via tauri-driver
docs/architecture/ Architecture overview
docs/adr/ Architecture Decision Records
docs/brand/ Logo, colors, and the social preview artwork
docs/development/ Setup, configuration, versioning
docs/discovery/ Repository presentation assets and validation
docs/images/ Screenshot contribution guide
docs/phases/ Phase checklists and acceptance reports
scripts/ Repository tooling
Further crates from the plan are added when the phase that needs them begins — see
ADR-004.
Release history
Every version, newest first, with its notes, known limits, and downloads, is on
GitHub Releases. The same notes are kept in
docs/releases, one file per version.
© 2026 8 West Ventures, LLC.
