Early version v0.1.0 · Native agents, one terminal workflow.Current status

Navigation

// Native agents. User-defined priorities.

One supervisor.
Your agents.
Their native
capabilities.

Your coding agents already know how to work.
uuldana-cli keeps them working together, through one terminal workflow, on your terms.

Early version. Agents installed on the machine are found by themselves and run through it. See what works.

UULDANA-CLI / WORKSPACEDESIGN ILLUSTRATION
One session, multiple native agentsYour priority
01
Agent Anative terminal · own configuration
ACTIVEQUOTA REACHED
↓ usage quota reached
↓ visible context + same permission ceiling
02
Agent Bnative terminal · own configuration
NEXT ELIGIBLEACTIVE
User permission ceiling Native plugins & MCP
Illustration: Agent A is active. Agent B is next in your priority order.

Illustrative agents, ordered by you. This is how a switch at a usage limit works.

Real native coding agents Provider-owned authentication Open-source architecture Provider-agnostic contracts
01 / The direction

The agent changes.
Your workflow shouldn't.

A supervisor for the boundaries between agents. Each capability links to what it does today and where its limits are.

01

Your agents. Your order.

It finds the agents you have installed. Choose their order and your accounts in one config file, or type /agent inside a session. uuldana-cli follows your order, and tries agents you did not rank after it.

Configuring agents
02

Keep going at the limit.

When a provider prints its usage limit, uuldana-cli stops it gracefully and continues on the next agent in your list. No Ctrl+C, no retyping, no relaunch.

Automatic failover
03

One thread of work.

A journal records every run of a session, and uuldana resume picks it up again. Work moves between agents as visible context, with honest fidelity labels.

Sessions & resume
04

Permissions stay deliberate.

You set a ceiling for approval, filesystem, and network. Switching agents keeps the same ceiling, and an agent that cannot enforce it is refused.

Permission behavior
05

Keep your native setup.

Agents retain their own authentication, plugins, skills, and MCP servers. Nothing is translated between agents, and what does not carry over is stated.

Plugin compatibility
06

Open at the boundaries.

Add a provider with a JSON manifest, or with a compiled adapter through the SDK. Provider-specific logic never enters the core.

Explore the Provider SDK
02 / A clear boundary

The real agent,
all the way through.

Whichever agent runs, you use that agent. uuldana-cli's architecture leaves reasoning, tools, and native behavior with it.

The agent's own terminal interface runs in a pseudo terminal. Your keyboard is relayed to it unchanged, and its screen is drawn under a header that stays at the top. Structured transports have different capabilities and are documented as such.

Understand the transport model
  • The agent does the coding.

    Its native tools, instruction discovery, skills, and agent loop stay inside the provider process.

  • The supervisor coordinates.

    Process lifecycle, provider-neutral events, session records, and policy checks belong to Uuldana.

  • You remain in control.

    Agent selection and priorities are yours. Native approvals and hidden agent state are not transferable context.

  • Compatibility is explicit.

    Shared plugins need support from the receiving agent. Provider-specific behavior stays provider-specific.

03 / Built in the open

A foundation,
with the status visible.

Contracts are frozen. The runtime, finding agents by themselves, sessions, and failover are in this early version. What was tried on real programs is listed.

Read the evidence and limitations
Core contracts & Provider SDKContract 0.2 frozen
Runtime, terminal relay & sessionsAvailable
Providers, failover & handoffAvailable
Linux & macOS: tests in CIPassing

Snapshot of the repository on 2026-10-31. The four agents installed on the test machine were found and run through it with no setup; limits were not tried again on a real exhausted account since the per-agent wording was removed.

// install

Install with npm, then run uuldana

For Windows x64. You need Node.js 18 or newer: it only runs a small launcher, and the program itself is a native binary.

sh
npm install -g uuldana-cli

Latest on npm: v0.1.0

  1. Install your coding agents the way you normally do. There is no list of agents and nothing to set up.
  2. Open a terminal in your project and run uuldana. The first start looks for agents, which takes a while; after that it starts at once.
  3. When an agent hits its usage limit, the work continues on the next one. uuldana resume picks the session up later.
// contact

Report a bug, share an idea, or get in touch

uuldana-cli is built by one person. Messages are read by that person, so say what you ran into and where, and keep keys and tokens out of it.

Criticism and suggestions

Something confusing, missing, or wrong in the tool or on this site. Short notes are fine.

Email feedback

Business and collaboration

Work together, a custom integration, or a question about using uuldana-cli in a team. Tell me who you are and what you need.

Email about business

Prefer a message on social media? Find me on Threads or Instagram. Bug reports are easier to follow by email or as a GitHub issue.

// support

Independently built, in free time

uuldana-cli is developed and maintained outside of a day job, completely free and open source. If it saves you time, a donation helps keep it moving forward.

On a personal note, I'm building these tools to strengthen my portfolio so I can pursue better opportunities abroad. The biggest reason behind this dream is my girlfriend in Kyrgyzstan, and we've been in a long-distance relationship between Indonesia and Kyrgyzstan for a while now. Every donation brings me one step closer to closing this distance and building our future together.

Open to collaboration and international opportunities. Bugs, criticism, suggestions, or business enquiries are welcome on the contact page.

Start with the first session.

Build it and run your agents through one terminal. Nothing to set up.

Read the documentation

Search documentation

Start typing to search all articles.

↑ ↓ navigate · Enter openEsc close