MULTIMODAL AGENT · ORIGINAL KERNEL · LOCAL-FIRST

Bring it anything.
Get finished work back.

EvoForge understands text, images, audio, video, documents and the live state of your computer. It connects models and tools to produce code, research, documents, images, voice, music and film, turning your material into work you can review, revise and keep.

V0.11.1Windows · macOSBring the models you trust · local or cloud
54original skills built in
98native host tools
5video generation backends
5voice synthesis backends
6interface languages
2desktop platforms
01 / ACTION + CODINGORIGINAL KERNEL · REAL ENVIRONMENTS

From intent
to verified work.

EvoForge works across real repositories, browsers, desktop apps, terminals and servers. Tools, context, memory, permissions and specialist agents stay connected, so multi-step work can move forward without losing its thread.

01

Connected tools, continuous work

Ninety-eight built-in tools read documents, work with browsers and desktop apps, run terminal tasks, and edit code by structure. EvoForge organizes context and attachments so long, multi-step work remains coherent.

02 LEADSWARM

Specialists working together

Complex work can be coordinated by a lead agent or shared across a team. Each specialist receives the tools and brief it needs, while progress and decisions remain visible in one place.

RED → GREEN 03

Test changes before they land

Code changes are checked in an isolated worktree. Tests reproduce the issue, confirm the fix, and remain as regression checks for future changes.

02 / EVERY MODALITYINPUT × OUTPUT · ONE SYSTEM

It sees. It hears. It reads.
Then it makes.

Work with text, images, audio, video and documents as first-class material. EvoForge understands the content, processes it with models and tools, and produces results ready for further use.

FIG.01 / MODALITY MATRIX · INPUT × OUTPUTTABLE · 5 ROWS
MODALITY IN — WHAT IT UNDERSTANDS OUT — WHAT IT PRODUCES
TextTEXT Conversation, files, web pages, your mail inbox, terminal output. Code, essays, reports, slides, replies sent from your own mailbox.
ImageIMAGE Photos, screenshots, the live screen of any desktop app. Generate and edit images, create icons and posters, and compress files to suit your quality requirements.
AudioAUDIO Speech and music understanding, transcription, word-level alignment. Voice from five providers, composed music, podcasts and audiobooks.
VideoVIDEO Understand video through its visuals and audio, from a file or a link. Create shots with five video generation services, add narration and subtitles, and export a finished video.
DocumentDOC PDF, Word, PowerPoint, spreadsheets, any format, any size. docx, pptx, xlsx you can hand to someone as is.
VIDEO GENERATION: VEO, GEMINI OMNI, SEEDANCE AND WAN · MUSIC: LYRIA · IMAGE: NANO BANANA
03 / THREE WORKSPACESWRITE · SEE · DIRECT

Different work needs
different places to think.

Chat starts the work. Three shared workspaces carry it to completion: a writing desk for publishable pieces, a canvas desk for visual composition and precise image edits, and a director's console for productions that unfold over time.

01 / WRITING DESK

From first draft
to publication.

Work with the agent on the same article: refine paragraphs, rearrange sections, plan illustrations and compare headlines. Return tomorrow and pick up exactly where you left off.

  • Edit paragraphs and sections in place
  • Plan figures before generating them
  • Keep title, prose, images and captions in one artifact
ARTICLE · NEWSLETTER · ANNOUNCEMENT · STORY
02 / CANVAS DESK

See the full picture.
Edit exactly where you choose.

Keep references, drafts, notes and alternatives together on one canvas for easy comparison. Mark the precise area to change, use other images as subject, style or scene references, and receive the result beside the original with an unchanged-pixel check.

  • Arrange source images and references spatially
  • Point, annotate, connect and compare
  • Edit a region while preserving everything outside it
IMAGE EDIT · COMPOSITION · CUTOUT · VISUAL EXPLORATION
03 / DIRECTOR'S CONSOLE

Bring picture and sound
together on the timeline.

Scripts, footage, generated shots, voice and music meet on a real multi-track timeline. Review every take, refine the edit, and keep the production visible from first plan to final export.

DIRECTOR'S CONSOLE / FULL TIMELINEFILM · PODCAST · AUDIOBOOK
FIG.02 / DIRECTOR'S CONSOLE · GENERAL ARRANGEMENTMONITOR · ASSETS · V3–V1 · A1–A3
MONITOR Margins back to 21%. Q3 · NARRATION 00:12 00:00:12:04 1920×1080 · 25p ASSETS REC · 01 GEN · 02 PLACEHOLDER GEN · 04 GEN · 05 · REVIEW REC · 06 STILL · 07 PLACEHOLDER NARRATION · 5 VOICES MUSIC · LYRIA FILL → REVIEW → REPLACE 0:000:060:120:180:24≈ 0:30 V3OVERLAY V2PICTURE V1BED A1NARRATION A2MUSIC A3SFX TITLE SUBTITLE · 2-FRAME RULE s01 · REC s02 s03 · FILL s04 · GEN · REVIEW s05 s06 · REC POSTER BED · LARGE TYPE, THREE SIZES DUCK -12 dB · s03..s05 s03..s05 EDIT TOGETHER WITH THE AGENT · UNDO EACH EDIT · EXPORT IN ONE STEP
  1. 01Set the script

    Shape the narration and structure before arranging the picture.

  2. 02Plan the shots

    Give every shot a clear place in the story and on the timeline.

  3. 03Generate and refine

    Create shots, voice and music, then refine them in context.

  4. 04Review

    Preview each take and choose what to keep, regenerate or remove.

  5. 05Export

    Export the finished production with subtitles and keep the edit list.

04 / SPECIALIST SKILLS54 BUILT IN · MORE CAN BE ADDED

One general agent.
Deep craft when the work demands it.

EvoForge comes with 54 original skills spanning finance, media, writing, systems, visual work and engineering. Each skill combines domain methods, references and practical scripts with task-specific guidance and quality checks.

01Finance

  • Stock verdict
  • Equity fundamentals + DCF
  • K-line structure (Chan, waves)
  • Stock research planning
  • DEX arbitrage design

02Film and media

  • AI film director
  • Narrated video director
  • Video ad director
  • Audio program director
  • Video analysis and recreation
  • Image generation craft
  • Icon design
  • Logo motion

03Writing

  • New story arcs for your novel
  • Outline restoration
  • WeChat article director
  • Natural, expressive writing
  • Deep research
  • Memory page

04Systems design

  • LLM pre-training framework
  • Post-training / RLHF framework
  • Local fine-tune engine
  • Robot sim2real
  • IM system design
  • AI gateway design
  • Social simulation
  • Companion agent

05Web and visual

  • three.js creation
  • Web 3D map engine
  • Game ability VFX
  • Text morphing
  • Website recreation
  • Boutique storefront
  • Motion and interface design
  • Algorithmic art

06Engineering and growth

  • Code review, four lenses
  • iOS app debug
  • Competitor growth teardown
  • Distill a design skill from repos
  • Skill creator and installer
  • Dynamic slides

EvoForge turns useful experience into memory, refined skills and new tools. Evaluation and approval decide which improvements carry forward into future work. Explore the evolution path →

05 / A DESKTOP BODYTHE WHOLE MACHINE · AND BEYOND IT

Your computer, connected as one workspace.

Move naturally between files, browsers, desktop apps, terminals and remote machines. Work follows your permissions, with important actions and results visible for review.

PL-01

Browser

Read pages, fill forms and complete online workflows, with permission checks and a record of each action.

PL-02

Any desktop app

It sees through the accessibility tree and screenshots, and acts with the mouse and keyboard. Windows and macOS, native or Electron.

PL-03

Shell and server fleet

Work with local and remote terminals in one workspace, preserving session state and following your configured sandbox policies.

>_
PL-04

Mail and channels

Keep long-running tasks moving, run scheduled work, and receive updates by email, push or desktop notification. New work can also arrive through your inbox.

ClaudeOpenAIGeminiDeepSeekQwenKimiOpenRouterOllama

Bring the work.
Keep the result.

Windows and macOS. Your models. Your files. Your control.