MyLaude 0.4.25

MyLaude is a local-first agentic coding CLI for people who want a sharper terminal workflow for coding, research, review, and multi-agent execution.

It is built to stay explicit where many agent systems get vague: file reads should be complete enough to trust, workflow escalation should be visible, safety modes should be named, and setup should not feel like unpaid integration work.

Why MyLaude

MyLaude exists to push against a few recurring problems in coding-agent loops:

• acting too early after only partial file reads
• retrying the same failure pattern until tokens disappear
• burying routing, context, and verification state
• making new users solve tooling friction before they can do useful work

The goal is a cleaner runtime, clearer workflows, and a friendlier path for people who are still learning how to work with AI systems inside real codebases.

What Changed In 0.4.25

• the interactive input box now uses a logical cursor, so left/right edits and long backspace sequences no longer tear the visible session UI
• typing / in interactive mode now opens slash-command suggestions in-place, with arrow navigation plus Tab or Enter completion
• /model:config and /mind now leave visible feedback in the active chat loop instead of disappearing into transient output
• onboarding and selection menus now treat Esc as go-back behavior instead of silently killing the whole interactive flow; Ctrl+C remains the real exit
• Telegram bridge now rolls over overgrown remote sessions, sanitizes outbound text, and mirrors both inbound and outbound Telegram messages in the terminal
• /context is now available in both local interactive mode and Telegram remote mode to inspect the current session snapshot and recent context
• the HUD now shows only real provider usage numbers; estimated context-limit bars have been removed
• English welcome text and Telegram auto-start warnings were cleaned up to remove encoding pollution and broken mixed-language separators
• BytePlus onboarding, Responses runtime, Gemini presets, and provider diagnostics from 0.4.24 remain in place

Core Workflows

• run for direct one-shot work
• niko for ordinary multi-agent planning, shaping, and synthesis
• athena for broader multi-agent execution on wider tasks
• contest for adversarial proposal, critique, and verdict
• design for frontend direction, UI critique, and implementation-ready polish briefs
• doublekill for guided dual-model setup with an execution model and a brain model

Built-In Capabilities

• version-aware docs lookup with docs
• selectable docs fallback backend with search-engine
• Gemini Deep Research as an independent long-form research engine
• Telegram as an optional remote control bridge
• MyLaude.MD as repo-level operating memory
• WordUP autosave snapshots
• interactive HUD with model, token, and context progress
• verification planning, evidence tracking, and contradiction surfacing
• explicit permission modes plus --whosyourdaddy for intentional accept-all

Install

Official npm install

Once published to npm, the intended install flow is:

npm install -g mylaude
mylaude --version
mylaude

Official tarball install

If you are installing from a direct release file instead of npm:

npm install -g .\mylaude-0.4.25.tgz
mylaude --version
mylaude

Source install

If you are developing locally:

cd <your-local-mylaude-repo>
npm install
npm run build
npm link
mylaude --version

First Launch

On first interactive launch, MyLaude guides the user through the basics:

• language selection for Chinese or English
• workspace trust confirmation before the main session starts
• execution API URL, model, and key
• local alias for the execution provider
• optional doublekill dual-model setup with an execution model and a brain model
• docs fallback choice when external lookup first needs a web backend
• optional Telegram setup if the user wants remote control

Gemini Deep Research is built in but optional. If a user chooses gemini-deep-research or runs deep-research before configuring it, MyLaude now offers inline setup and then continues the query.

Provider onboarding now includes preset flows for:

• OpenAI
• Anthropic
• Google Gemini
• BytePlus
• OpenRouter
• Ollama
• Custom provider

BytePlus is now split into:

• BytePlus > Coding for coding-oriented models such as ark-code-latest, bytedance-seed-code, kimi-k2.5, glm-4.7, and similar Coding Plan models
• BytePlus > Chat / Reasoning for general chat, reasoning, and multimodal chat models on the standard /api/v3 path

BytePlus > Responses + MCP now uses a dedicated experimental Responses runtime with provider-native function tool calling. Local MyLaude tools can execute through that path; remote MCP remains experimental.

Google Gemini now suggests:

• gemini-3-pro-preview
• gemini-3-flash-preview
• gemini-2.5-pro
• gemini-2.5-flash

Quick Start

mylaude
mylaude run "make this repo easier to understand"
mylaude athena "finish the remaining runtime features"
mylaude niko "plan the next CLI improvement"
mylaude doublekill
mylaude design "improve the landing page hierarchy"
mylaude contest "should we refactor the tool runtime now?"

Docs and Research

mylaude docs "react 19 useActionState" --docs-library react --docs-version 19
mylaude search-engine google
mylaude search-engine bing
mylaude research-engine gemini-deep-research
mylaude deep-research-config
mylaude deep-research "compare the best approaches for long-context repo memory"

Use docs for targeted references. Use deep-research for broader synthesis.

Memory and Autosave

Project memory:

mylaude mylaude-md
mylaude revise-mylaude-md <session-id>
mylaude revise-mylaude-md <session-id> --apply

Progress snapshots:

mylaude wordup
mylaude wordupnow

MyLaude.MD holds repo-level commands, architecture notes, workflow rules, gotchas, and durable learnings. WordUP protects work against interruption by writing markdown snapshots of the current state.

Telegram

Telegram is optional and user-configured:

mylaude telegram

The setup flow can remember whether Telegram should auto-start on future CLI launches.

Once linked, Telegram can keep controlling the same local CLI session without requiring an extra plugin install.

Inside interactive mode you can also use:

• /model to switch the current execution model
• /model:config to reopen the API setup wizard for the execution model
• /mind to swap the execution model and brain model immediately

Diagnostics

mylaude doctor
mylaude doctor --test-providers
mylaude providers
mylaude sessions
mylaude evidence
mylaude conflicts
mylaude verify

Interactive Slash Commands

• /athena <task>
• /design <task>
• /niko <task>
• /contest <task>
• /language
• /docs <query>
• /search-engine [bing|google]
• /research-engine [builtin|gemini-deep-research]
• /deep-research <query>
• /deep-research-config
• /doublekill
• /wordup
• /wordupnow
• /mylaude-md
• /revise-mylaude-md [--apply]
• /doctor [test|--test-providers]
• /sponsor
• /mode <prompt|read-only|accept-edits|accept-all>

Safety Model

MyLaude keeps permission modes explicit:

• prompt
• read-only
• accept-edits
• accept-all

--whosyourdaddy maps directly to accept-all.

Build and Release

npm run build
npm run typecheck
npm run test:smoke
npm pack

The release tarball ships from compiled runtime output and excludes local session state, local provider secrets, and workspace-specific data.

Product Notes

• MyLaude keeps the terminal workflow explicit: routing, verification, and workflow escalation should stay visible.
• MyLaude tries to reduce common coding-agent failure modes such as acting on partial file reads, retrying the same broken path, and hiding setup friction inside vague magic.
• The goal is not to feel minimal in screenshots. The goal is to feel faster, clearer, and less wasteful when real work is happening.

Support

mylaude sponsor

Sponsorship is optional and exists only to fund compute, testing, and further upgrades.