The agentic dev environment
for local LLMs
Build with one agent or automate repeatable work with a team. Ascora ADE reads your code, searches the web, edits files, runs commands, and keeps its tools and changes visible — powered by local models, CLIs, APIs, or the web chats you already use.
LM Studio · Ollama · OpenRouter · OmniRoute · Codex · Copilot · Claude Code · Gemini CLI · GLM — or signed-in ChatGPT, Claude, Gemini, Grok, DeepSeek, Qwen, Alice & Mistral web chats.
One agent, every backend — local models, CLIs, or the web chats you already use
LM Studio
Local · OpenAI-compatibleOllama
Local · OpenAI-compatibleOpenRouter
Cloud API · many modelsOmniRoute
Managed gateway · 256 providersCodex CLI
OpenAI · sandbox controlsGitHub Copilot CLI
Device login · reasoningClaude Code
Opus · Sonnet · HaikuGemini CLI
Google · geminiGLM / ZCode
Zhipu & friendsAscora WProvider
Web chats · no API keyDesign the work once. Let a team run it.
Blueprints turn several WProvider agents and practical actions into a reusable visual workflow. Branch on results, wait for branched work to converge, repeat a review loop, then deliver the outcome — manually or on a schedule.
Start
Research release
web_search · web_fetch
Inspect workspace
file · search · command
Review & decide
fan-in · RESULT=PASS
Publish via HTTP
POST · response body
Notify the team
Telegram · webhook
Fix and re-check
bounded feedback path
Graph routing, not a rigid chain
Fan out to every matching route, fan in after every predecessor finishes, and branch on success, failure, fallback, contains, or equality. Organize the canvas with sticky notes, pan, zoom, and fit-to-view.
Visual canvas · Graph or List
Bounded agents and feedback loops
Enable agent mode globally or per step for a bounded tool loop. A dedicated Repeat input creates a controlled reviewer/fixer path with 2–20 passes.
web_search · web_fetch · project tools
Run, schedule, audit, and connect
Start manually or at a minute interval, follow every action and tool call in the live log, or connect several scenarios to one shared prompt in Blueprint Hub.
Hub · Telegram · HTTP · webhooks
A full development environment, with an agent inside it
It edits the code — you stay in control
The built-in agent loop reads, plans, edits, runs, and verifies. In Ask mode it pauses on proposed writes and commands so you can inspect the inline diff or output; Auto-apply continues without each prompt. CLI agents follow their selected sandbox or permission profile.
- Reviewable inline diffs in the built-in tool loop
- “Ask before changes” or “Auto-apply” for the built-in loop
- Provider-specific sandbox and permission controls for CLI agents
- A running tally of every file the task touched
src/theme.ts+12-3needs reviewPrivate by default, on the models you choose
Ascora ADE speaks to local LLMs through the OpenAI-compatible APIs of LM Studio and Ollama, and to hosted models through OpenRouter. Prefer a CLI agent? Codex, GitHub Copilot, Claude Code, Gemini CLI, and GLM/ZCode all plug in. Pick a backend per chat; network-capable providers and tools are always an explicit choice.
- Local model requests stay on your machine with LM Studio or Ollama
- OpenRouter plus Codex, Copilot, Claude, Gemini, and GLM CLIs
- Each chat remembers its own model — pin one per task
qwen2.5-coder-7bllama3.1openrouter/autosonnetgeminichatgpt · claudeOne API key, your OpenRouter model catalog
Enable OpenRouter in Agent backend settings, save the API key locally, refresh the available models, and pin the right one to each task. Ascora ADE keeps the same provider and model across the coding loop, Git helpers, and usage analytics.
- Saved API key with normalization for Bearer and Authorization values
- Live model refresh, streaming chat, and clear connection diagnostics
- Per-task model pinning with openrouter/free available by default
sk-or-v1-••••••••••••••••Saved locallyProtect the connection without leaving your IDE
Ascora ADE 1.3.0 adds a built-in VPN workspace with live public IP and country, server health, load, measured latency, protocol choice, and traffic usage. Windows builds bundle both VPN engines; Linux uses the system WireGuard or OpenVPN packages.
- WireGuard and OpenVPN are bundled on Windows — no separate install
- Automatic best-server mode, plus manual server selection on Premium
- Anonymous access includes a locally enforced 100 MB daily allowance
Turn supported web chats into coding agents — no API key
WProvider drives a supported provider’s web chat as an agent backend inside a hidden browser window. Sign in once and Ascora ADE types each turn into the site’s composer, reads the reply back, and runs tools on your project. Usage remains subject to that service’s plan, limits, and terms.
- Use ChatGPT, Claude, Gemini, Grok, DeepSeek, Qwen, Alice & Mistral as agents
- Sign in once; the session persists in its own partition
- Full tool use and streaming — even works over SSH
chatgpt.comclaude.aigemini.google.comgrok.comchat.deepseek.comSearch the codebase — and the web
Follow the agent’s thought process while it reads files, searches the workspace, reaches the web, edits code, and runs commands. Each call appears in a visible tool card; command cards show captured output and exit status when the process finishes.
- Read-only web_search and web_fetch with bounded page text
- Ripgrep-style regex, glob, context, and case-sensitive search
- Tool cards with output, diffs, exit codes, tokens, and time
prefers-color-scheme accessibility4 resultsuseTheme · **/*.tsx · context 29 matchesnpm testdoneThemeToggle and wired it to persisted settings. Tests pass.A project shell, right where you work
Run ordinary shell commands, builds, tests, and scripts without leaving the workspace. The xterm.js interface streams output from the platform shell, while the agent can run commands and report their exit status too.
- Interactive line-oriented shell inside the workspace
- Copy with Ctrl/Cmd+C or right-click
- No full-screen TUI or mid-command stdin support
Read every change, not a wall of diff text
A source-control panel lives beside the editor: stage files, inspect a Monaco-powered syntax-highlighted diff, switch branches, commit, and push without a separate app. Let the agent read your staged changes and draft the commit message for you.
- Language-aware unified diffs with old/new line gutters
- Per-file +added / −removed stats and binary badges
- AI-written commit messages from your staged diff
- Readable split layout with the file list always in view
feat/theme-toggle3 stagedMonaco editor
The same editor that powers VS Code, bundled and fully offline.
SSH terminals
Drive remote hosts from the same agent loop over SSH — each host keeps its own chats.
Background tasks
Agent runs keep progressing when you switch tasks, and restore their live state.
In-app updates
An Update badge opens the full changelog and a one-click download.
Live preview
Built-in live server renders your HTML as you change it.
Analytics that explain usage
Estimate API cost, follow model share over 12 weeks, and spot work patterns on an hour heatmap.
Six interface languages
English, Russian, Ukrainian, German, French, and Italian are available from the rail.
Attach files & images
Drop files or images into a prompt and the agent works with them directly.
A model per chat
Every task pins its own model and provider, restored when you reopen it.
Drag-and-drop files
Reorganize the Explorer tree by dragging files and folders into place.
Your code, your machine
Local model requests stay local; hosted providers, web tools, and network Blueprint actions are explicit choices.
Dockable panels
A VS Code-style layout you can split, stack, and re-dock.
Web research tools
Search the web and fetch readable page text without handing an agent a full browser.
Search that scales
Combine regex, globs, case matching, and surrounding context across local or SSH workspaces.
Archive, inspect, restore
Open archived conversations read-only, then restore the task with one click when you need it.
Workspace names, your way
Rename a project inline from the rail without changing its underlying folder path.
Four steps, no context switching
Open a project
Point Ascora ADE at any folder. It indexes the tree and connects to your chosen model.
Describe a task
Ask in plain language — “add a dark-mode toggle”, “fix the failing test”, “refactor this”.
Review the diff
The agent plans and proposes edits. Inspect each change as a clear, reviewable diff.
Apply with confidence
Approve to write to disk, or let Auto-apply run the whole task end to end.
Start building with your own models
Free to download. Install Ascora ADE, point it at a project, and ship your first AI-assisted change in minutes.
view the changelogFrequently asked
With LM Studio or Ollama, the model and normal workspace tools run locally. Network traffic is explicit: hosted and CLI providers call their services, WProvider drives the selected web chat, web_search/web_fetch reach the internet, and Blueprint HTTP, Telegram, or webhook actions contact the destination you configure.
Use OpenAI-compatible local models through LM Studio or Ollama, hosted models through OpenRouter, or the Codex, GitHub Copilot, Claude Code, Gemini, and GLM/ZCode CLIs. Ascora WProvider also drives supported signed-in web chats. The backend and model are pinned per task.
Yes. Ascora WProvider drives a supported provider’s web chat inside a hidden browser window, so you can use ChatGPT, Claude, Gemini, Grok, DeepSeek, Qwen, Alice, or Mistral without configuring an API key. Sign in to each service once; its session persists in a dedicated browser partition. The provider’s own plan, limits, and terms still apply.
A Blueprint is a reusable automation scenario. Several named WProvider agents can collaborate through a visual graph with conditional fan-out, fan-in, and one bounded 2–20-pass Repeat path. Mix agent steps with file, command, HTTP, Telegram, webhook, and delay actions, then run manually or on an interval while ADE is open.
Yes. Agent mode can be enabled globally or per step for a bounded multi-turn loop. web_search and web_fetch work without a project; selecting a local project also enables file, search, edit, write, and shell tools. Tool calls and their results are kept in the execution log.
The language switcher includes English, Russian, Ukrainian, German, French, and Italian. Locale-aware screens such as Analytics follow the selected language, and untranslated Ukrainian strings fall back to English.
Ascora ADE is built around local-first, agentic workflows: it’s designed to run against models you host yourself, with a transparent read-plan-edit-run loop and explicit diff approval. You own the stack end to end.
The built-in agent loop can use “Ask before changes” to wait for approval on writes and commands, or Auto-apply to run uninterrupted. CLI agents follow their selected sandbox or permission mode. Blueprint project tools are autonomous when you enable agent mode and bind a project, so review the graph and workspace before running it.
Current 1.3.0 builds target Windows x64 and Linux x64. Windows is available as an installer and portable build; Linux packages include AppImage, .deb, .rpm, and .pacman. A macOS build is not currently published. The browser demo uses mock data.
Yes. Version 1.3.0 adds Wandrounik VPN with WireGuard and OpenVPN, live IP and country status, automatic server selection, and traffic usage. Windows builds bundle both VPN engines; Linux uses the corresponding distro packages. Anonymous use includes a locally enforced 100 MB daily allowance.
Yes. The Ascora ADE source is available on GitHub under the MIT license.