Coming to Fab — the Unreal Engine marketplace. One-time license, no subscription. Be notified at launch →
Studio · Unreal Engine tooling

The AI copilot
that sees your game.

Sentari AI Assistant docks into Unreal Engine as a panel you talk to. It captures your viewport, understands the scene, and builds alongside you — laying out Blueprints, scattering PCG worlds, tuning Gameplay Abilities, crafting materials and VFX, all by calling the real editor tools.

265+verified tools
5.7 + 5.8Unreal Engine
Offline-readylocal or cloud
sentari · live session
prompt viewport blueprint Sentari reason spawn scatter wire
Why Sentari

Detailed enough for real work. Easy enough to feel like magic.

Most "AI for game dev" lives in a browser tab, far from the engine, charging you per token to guess. Sentari is the opposite: a teammate that shares your workspace, works offline on free models, and only ever touches your project through verified, non-destructive tools.

Detailed, yet non-destructive

Node-level depth — individual pins, attributes and graph edges — wrapped in a workflow that verifies every change before it lands. Every action is undoable; nothing silently overwrites your work.

verify · compile · undo

Ask it to look, and it only looks

"Why is this scene so dark?" is a question, not an instruction. Sentari tells the two apart — in whatever language you ask — and runs inspection requests with the write tools physically out of reach. It reports what it found, offers the fix, and waits for your yes. Even when approvals are set to automatic.

inspect-only · then it asks

Local models, built in natively

LM Studio and Ollama run inside the plugin — not through a flaky side-channel. Native tool-calling, VRAM-aware keep-alive, and per-model capability probing, all wired straight into the panel.

ollama · lm studio · native

Any model, free or frontier

Bring your own. Route to free open-source weights locally, or reach any frontier model through OpenRouter and friends. Start on a free Llama or Qwen; scale to Claude or GLM when the task demands.

byom · openrouter · api

Compact or extended, on demand

One dockable panel that shrinks to a slim chat rail when you're focused, and expands into a full workspace — chat, tools, 3D preview and logs — when you need the whole bench. Your layout, your call.

dockable · resizable

Offline by design

Run fully offline — your project never has to leave the machine. The same tight, local-first architecture that makes it private is exactly what makes it cheap: no per-prompt tax, no account to feed.

offline mode · private

A genuine budget saver

Keep routine, repetitive work on a free local model and spend your cloud budget only on the hard reasoning. The money stays in your pocket — one plugin license, no subscription, no credits to burn.

no subscription · no credits
Token & budget saver

Run the routine locally. Spend cloud only where it counts.

Sentari lets you split the workload the way a smart team does. Free local models handle the volume — spawning, renaming, wiring, audits — while your paid cloud model is reserved for the heavy lifting. You stay fast, and your bill stays small.

Lane 01 · Local · free

The daily grind, on free weights

High-volume, low-risk work runs on a local Llama, Qwen or Gemma — zero per-prompt cost, fully offline.

  • Scatter & place actors
  • Rename, organise, audit assets
  • Wire nodes, set pin defaults
  • Read attributes & scene state
Sentari routes
the task
Lane 02 · Cloud · on demand

The hard parts, when you need them

Complex reasoning, vision and long planning go to a frontier model — invoked only when the local lane can't carry it.

  • Design a system from scratch
  • Read the viewport (VLM)
  • Debug a tricky graph
  • Long multi-step builds
Vision

Give it a photo. Get a lit scene.

One prompt, one reference image, and a local model on an ordinary GPU — sun angle, colour temperature, height fog, skylight bounce and a post-process pass, all set through the editor's real tools and all undoable.

“Set the atmosphere and lighting like in the picture I gave you.”

The Sentari panel with a sunset reference photo attached; the local model lists the exact lighting changes it made, with the result filling the viewport
The turn

Reference in, itemised changes out — from a local model, offline.

The resulting scene from the player's point of view: a warm low sun, orange haze and long shadows
The result

The same scene in play. Nothing hand-tuned.

See the full breakdown

Bring your own agent

Claude Code, Codex, or any agent — running inside the editor.

Sentari isn't a walled garden. A real terminal lives inside the panel, and the same MCP bridge the assistant uses is open to any agent you trust. Launch Claude Code or Codex with one click, or point any MCP-speaking client at it — and it drives the engine through the exact same verified, non-destructive tools.

Claude Code · one-click launch OpenAI Codex · one-click launch Any MCP client · connect to :8765 Custom agent · terminal + tools
sentari · agent lane
# launch Claude Code inside the panel
$ claude
› claude code connected · mcp bridge :8765

# it now drives the editor through the same tools
 "build a campfire with emissive embers and a crackle loop"

  build_material_graph  →  master + instance, emissive on
  spawn_actor           →  SM_Campfire at the cursor
  generate_sfx          →  crackle loop → SoundWave, placed

✓ done 3 tools · verified · undo available

One bridge, every brain.

The MCP bridge that powers Sentari is a first-class interface for anything that speaks the protocol. Your favourite agent gets the same reflection-checked, verify-before-claim toolset — no re-wrapping, no second API to learn.

  1. Open the panel. The terminal and MCP server are already listening on :8765.
  2. Launch an agent. Claude Code and Codex start from the in-panel launcher; anything else connects straight to the bridge.
  3. Let it build. Every call flows through the same non-destructive tools and guardrails as the chat — identical safety, a different brain at the wheel.
The studio

A focused engineering shop, not a feature factory.

We build a small number of tools and build them deeply. Each one is shaped by months of daily use inside real production — measured against whether it genuinely removes friction for the person at the keyboard, not whether it demos well.

01 / Flagship product

Sentari AI Assistant

An AI collaborator that lives inside the Unreal editor. It reads the scene, authors Blueprint graphs, builds materials, scatters PCG worlds, drives Gameplay Ability System data, generates 3D & audio, and reasons about what's on screen — through a verified, node-level toolset.

sentaridigital.com / Sentari_AI_Assistant →
02 / Engineering

Editor-native systems

Performance profiling, asset auditing, and workflow automation built against the engine's real Python and C++ surfaces — no shallow wrappers.

03 / Consulting

Real-time R&D

Selective engagements for studios that need a hard problem actually solved: tooling architecture, GAS pipelines, PCG systems at scale.

How we work

Principles over polish.

P · 01

Verify, then claim

Every capability is built against the engine's real API and runtime-tested before it ships. No hallucinated tools, no guessed behaviour.

P · 02

Node-level depth

We go past "it works" to the individual pin, attribute, and graph edge. Depth is the product.

P · 03

Built for local, ready for cloud

Tuned to run well on modest local hardware — offline, your project never leaves the machine. That same tight design makes it fly on cloud models.

P · 04

Honest ownership

One-time license, no subscription theater. Full source, documented, and yours to keep.

265+
verified editor tools
5.7+
Unreal Engine versions
:8765
MCP bridge
100%
offline-capable
Coming to Fab

Building something in Unreal?

Sentari AI Assistant launches on Fab with a one-time license — no subscription, no credits. Drop your email and we'll let you know the moment it's live, or tell us about a hard real-time problem you'd like a hand with.

Notify me at launch See the plugin