v1.5 Release Notes

This release is about what happens after the answer. v1.5 adds a real document engine so the things your agents produce are editable artifacts instead of throwaway files, and a way to chain agents together so multi-step work runs on its own.

Arkios Studio: a native editor for presentations, documents, and designs, built and edited by AI or by hand.
Workflows: connect agents into a dependency graph, run it on a schedule, and let each step feed the next.

Arkios Studio

Until now, a deck or report from Arkios came back as a file. You could read it and download it, but changing one line meant asking for the whole thing again. Studio replaces that with a structured document you can open, edit, and keep.

Everything in a Studio document is an object the editor understands, so both you and the AI are working on the same thing. Ask for a change in chat and the slide updates in place. Drag the same element yourself and it moves. There is no regeneration step, and no file to re-download.

Three kinds of document

Deck. Presentations on a 1280×720 canvas. The AI composes each slide as a full design: a background colour or gradient, an accent band, a kicker, a heading, supporting copy, and icons or charts where they earn their place. Four themes ship in the box (arkios-light, arkios-dark, midnight, boardroom), and Present mode runs the deck full screen.

Doc. Paginated text documents with real page setup, letter or A4, with margins. Start from a template (resume, report, memo, letter, or blank) and build with structured blocks: headings, paragraphs, bullet and numbered lists with nesting, checklists, callouts, tables, dividers, and two-sided rows for things like a job title against its dates.

Design. Fixed-size artboards for anything that is a graphic rather than a page: a 1200×628 web banner, a 1080×1080 social post, a 1080×1920 story, a 1500×500 header. You set the dimensions and compose freely.

Working in the editor

Studio gives you direct control over every element, not just a preview:

Canvas and inspector: position, size, rotation, opacity, fills, gradients, strokes, shadows and glows, and per-image filters.
Layers and filmstrip: reorder elements by depth, and reorder slides by dragging.
Rich text: bold, italic, underline, strikethrough, colour, size, and links as inline marks, with a formatting toolbar in place.
Charts as vectors: data visuals are inserted as SVG, so they stay crisp at any zoom and can be edited rather than re-rendered.
Icons and images: a searchable icon picker, plus image masks (circle, squircle, hexagon, blob, and more) for framing photos.
Revision history: every change is a numbered revision you can review and restore.

AI that edits, not just generates

The AI panel inside Studio is connected to the open document. It reads the current structure before it changes anything, then applies a batch of operations, so edits are surgical rather than a full rewrite. It can reword a heading, restyle an element, add or redesign a whole slide, generate a hero image, or edit an image already on the page, including background removal, recolouring, and restyling, and swap it in place.

Ask for a document in ordinary chat and the AI builds it as a Studio document with real content in it. Open it in Studio when you want to iterate.

Export

Deck: PDF of all slides, PNG of the current slide, or a ZIP of every slide as PNG.
Doc: PDF of all pages.
Design: PNG or JPG of the current artboard, or PDF.

Workflows

Agents are good at one job. Real work is usually several jobs in order, where the output of one feeds the next. Workflows let you wire agents together into a graph and run the whole thing as a unit.

Build the graph

The workflow builder is a canvas. Drop a step, pick the agent that runs it, write the prompt, then draw a connection to say what has to finish first. Steps with no dependency between them run in parallel; a step with two upstream dependencies waits for both. Cycles are rejected as you draw them, so a workflow can never deadlock. A workflow can hold up to 20 steps.

Each step receives the output of the steps it depends on as context, so a research step can hand findings to a drafting step, which hands the draft to a reviewer.

Run it, or let it run itself

On demand: trigger a run and watch it progress live.
On a schedule: attach a cron expression for recurring runs, or schedule a single run for a specific time.
Via API: trigger a run from outside Arkios.

Every run streams its state as it happens: which steps are queued, running, done, failed, or skipped, with per-step output you can expand. Run history is kept, so you can open any past run and see exactly what each step did.

Handling failure

Steps are not all equally critical, so each one carries its own failure policy:

Fail workflow: stop everything.
Retry then fail: retry, and stop if it still fails.
Retry then continue: retry, and carry on regardless.
Continue: move on without retrying.

Each step also has an iteration cap, so a step that gets stuck in a tool loop is bounded rather than running indefinitely. Runs can be cancelled mid-flight.

Workflows that learn

When a step fails for a behavioural reason, such as a bad assumption, a missing instruction, or the wrong tool, Arkios distils what went wrong into a lesson and carries it into later runs of that workflow. Infrastructure failures like timeouts and rate limits are deliberately excluded, since they say nothing about how the agent should behave. Over time a scheduled workflow gets more reliable at the specific job you gave it.

You can also get notified on completion, either every time or only when a condition you define is met.

What This Means for Your Team

Decks, reports, resumes, and marketing graphics come out of Arkios as documents you can actually edit and export
The AI works on the same document you do, so revisions are a conversation rather than a regeneration
Multi-step work that used to be a person shuttling output between agents now runs as one scheduled workflow
Failures are handled explicitly per step, and recurring workflows improve as they run

Looking Ahead

We are continuing to expand Studio with more templates, brand controls, and export options, and deepening Workflows with richer branching and more trigger types.