Windows desktop · plain language · works out of the box
Get work done with AI
without learning to code
Claude Code, Codex CLI, Hermes, OpenClaw — powerful tools that normally expect you to live in a terminal. Aiglade puts them behind one guided window. Describe what you want; it takes it from there.
Aiglade is a Windows desktop application that runs multiple command-line AI agents at once. You describe a task in one place; Aiglade dispatches it to a suitable agent and queues it. Actions that could damage something stop and wait for your approval instead of running straight through. Aiglade uses the FreeModel model gateway by default, and you can point it at any other endpoint.
What AI can actually do for you
Nothing exotic. These are things today's AI already handles well — the hard part has always been getting it started.
Write and rewrite
Draft something, tighten the wording, change the tone, translate a passage, turn scattered notes into a finished piece.
Look things up and compare
Find out what something is, weigh a few options against each other, condense a pile of search results into the points that matter.
Sort out files and data
Merge spreadsheets, rename a folder by rule, pull fields out of documents, tidy up messy data.
Help with programming
Decode an error message, add a small feature, automate something you keep doing by hand.
Watch something on a schedule
Have it check on something at set times and hand you a short summary of what changed.
Run several jobs at once
One AI per job, several jobs in flight together, instead of waiting in a line.
Why it's not hard to use
AI tools usually fail people at the door, not at the task. Aiglade takes out most of that door.
Just describe it
No commands to memorise, nothing to script. Say what you want in ordinary language.
Walked through the first run
The app leads you from install to your first finished task. No documentation reading required first.
It asks before it breaks things
Deleting files or changing important settings stops and waits for your go-ahead. Don't like a step? Send it back.
Interrupted work picks up again
Laptop closed, app crashed, connection dropped — the job isn't lost. Reopen and it continues where it stopped.
Everything is on the record
Wondering days later which file changed, or who approved what? Look it up instead of relying on memory.
Runs on your machine
Not another website. Tasks, records and approvals stay on your computer; only the actual model call goes out.
Start here: AI agents
New to the term? These three explain the basics without assuming you write code.
What is an AI agent
How it differs from a chat model, and why it can act instead of only answering.
BasicsAgent vs ChatGPT
Same underlying models, different hands — where the difference actually shows up.
BasicsDo you need to code
What you can do with a coding background, and what still works without one.
Aiglade guides
From install to a working setup, including the parts people get stuck on.
Install and run your first task
Install, pick an agent, dispatch a task, watch it finish.
GuideSplitting work across agents
Which tasks parallelise cleanly, and which ones trip over each other.
GuideWhat the approval gate is for
Which actions get held, and how to configure it so it stops nagging.
If you want the technical version
The sections above are written for users. This one is for people who want to know what the middle layer actually does.
Tasks form a dependency graph, not a queue
Dependencies between subtasks are stored as a directed graph; completing an upstream node activates its downstream events. Cycles are detected. Six workflow templates ship built in (software delivery, bug fix, content pipeline, monitor brief, intent routing, quality optimisation) across six node types — start / step / gate / route / evaluate / end — and you can wire your own on the canvas.
ApprovalThe approval gate is persisted, not a dialog
Pending state is written to the database, so timeout countdowns survive a restart. Three triggers: workflow gate nodes, 13 built-in rules for dangerous commands, and plugin publishing. Four decisions — approve, reject, modify-then-release, or let it expire. Reject is not a dead end: it loops back and re-runs the previous step.
Three tables merge into one timeline
Task activity, workflow events and every agent tool call land in separate tables, then merge per task into a full chain: who dispatched it, which candidate was skipped, and why. Most multi-agent tools are one-shot sessions that leave nothing behind — this is where Aiglade differs most.
Crashes are classified, not just logged
Progress is checkpointed on a throttle. After an abnormal exit the app sweeps the queue: checkpoints present means resumable, absent means failed. Resuming feeds the partial output back in to continue. Three consecutive model failures trip a circuit breaker instead of retrying forever.
By the numbers: 24 tables in the local database, 131 HTTP endpoints, concurrency defaults to 10 tasks (5 per agent), 8 agent nodes seeded.
Which AI it connects to
Four command-line agents today, ready once installed. Model credit behind them defaults to our own FreeModel gateway, and can be pointed elsewhere.
Claude Code
Anthropic's command-line agent.
Codex CLI
OpenAI's command-line agent.
Hermes
Open-source agent runtime.
OpenClaw
Used for conversational terminals.
Choosing a platform
Agent tools split into a few categories that suit very different situations. Get the category right first.