AI render wrangler

An AI render wrangler for your own farm

Connect Claude, Codex, Cursor or any MCP client to RenderFlow. Ask why a frame failed, requeue it, or plan the night machines, straight from the chat. It asks before it changes anything.

Try RenderFlow

Try RenderFlow

Book a demo

Book a demo

Works with

Claude
Codex
Cursor
VS Code
LM Studio

and any MCP client

Claude
Which jobs failed tonight, and why?
renderflowfind_jobs{"status":"failed"}
total: 1 · Spot_30s_shot04 · status: failed · progress: 73
renderflowfind_tasks{"job_id":"671f…c204","status":"failed"}
total: 1 · index: 4 · status: failed
renderflowget_task_log{"id":"6720…9e1b"}
known_failures: cinema-0004 · "Cinema 4D could not write the output"
One job failed: Spot_30s_shot04. Its task for frames 41 to 50 stopped on RN-07
with "GeOutString: Files cannot be written - please check output paths!"
Your error catalog says: check the output folder exists and this machine can
write to it. The output is on N:, a mapped drive the RenderFlow service may
not see. Add N: under Settings → Mapped Drives and I can rerun the frames.
Done, N: is mapped. Rerun the failed frames.
renderflowcontrol_tasks{"ids":["6720…9e1b"],"status":"pending"}
message: "1 task(s) reset to pending"
Frames 41 to 50 of Spot_30s_shot04 are back in the queue.
Now put all the workstations to work.
renderflowfind_nodes{"type":"workstation","status":"suspended"}
total: 8 · WS-ANNA, WS-BEN, WS-CARL … · status: suspended
renderflowcontrol_nodes{"ids":["66f1…a3c0", …],"action":"resume"}
ids: 8 · applied: { status: "idle" }
All 8 workstations are back in the farm and available for work.
They pick up the queued frames of Spot_30s_shot04 right away.
Did any frames of Hero_CPU_v07 come out black?
renderflowfind_tasks{"job_id":"6721…3a07","status":"completed"}
total: 34 · index: 12, 21, 29 … · status: completed
renderflowget_task_preview{"id":"6721…4c12"}
image/jpeg
renderflowget_task_preview{"id":"6721…4c1b"}
image/jpeg
renderflowget_task_preview{"id":"6721…4c23"}
image/jpeg
I looked at three finished frames. Task 29 came out black; the other two
look right. Want me to put task 29 back in the queue?
Reply to Claude…

From question to fixed frame

Ask the farm in plain words

Render nodes by day, workstations at night, and your own cloud when you need more.

Logs for Spot_30s_shot04

Ask why a frame failed

It reads the task’s log with what your error catalog says about it, and tells you the fix. Where the catalog has no answer, it works from the log and says which parts it’s sure of.

Check the frames, not just the status

It looks at finished frames and points out the black or broken ones before anyone opens the folder.

Hero_CPU_v07
1001Failed
1002Failed
1003Failed
1004Failed
1005Failed
1006Failed
1007Failed
1008Failed
1009Failed
1010Failed
1011Failed
1012Failed
1013Failed
1014Failed
1015Failed
1016Failed
1017Failed
1018Failed
1019Failed
1020Failed
1021Failed
1022Failed
1023Failed
1024Failed
1025Failed
1026Failed
1027Failed
1028Failed
1029Failed
1030Failed
1031Failed
1032Failed
1033Failed
1034Failed
1035Failed
1036Failed
0%
Schedule
MonMonTueTueWedWedThuThuFriFriSatSatSunSun
00:0002:0004:0006:0008:0010:0012:0014:0016:0018:0020:0022:00
•02:00Run Script•08:00Change Pool•19:00Change Pool•02:00Run Script•08:00Change Pool•19:00Change Pool•02:00Run Script•08:00Change Pool•19:00Change Pool•02:00Run Script•08:00Change Pool•19:00Change Pool•02:00Run Script•08:00Change Pool•19:00Change Pool•07:00Wake•20:00 - 08:00Suspend•20:00 - 08:00Suspend•06:00Shutdown
Mon 07:00 · Wake12 workstations woken
Mon 19:00 · Change Pool12 workstations moved to Render
Tue 08:00 · Change Pool12 workstations moved to Artists
Wed 02:00 · Run ScriptScript finished on 8 nodes
Wed 19:00 · Change Pool12 workstations moved to Render
Thu 08:00 · Change Pool12 workstations moved to Artists
Fri 02:00 · Run ScriptScript finished on 8 nodes
Fri 20:00 · SuspendPreview: 6 machines suspended
Sat 08:00 · ActivatePreview: 6 machines activated
Sun 06:00 · Shutdown12 workstations shut down

Change the plan from the chat

Requeue the failed frames, push the hero shot to the top, or take the night machines off at 8 AM. It shows you what it will do and waits for your yes.

Try it with your AI client

Try it with your AI client

What you can ask

Admin-controlled

A working day, asked in plain words

You decide what it can do

Why a frame failed, what to requeue, what to render tonight: ask the farm the way you would ask a colleague. Click one to read the whole conversation.

AI clients are off on a new farm. An admin turns them on, and decides who else may connect their own assistant.

Claudeon the RenderFlow farm
“Why did Spot_30s_shot04 fail last night?”Reads the log against your error catalog and names the fix: the output drive the service cannot see.
find_jobsfind_tasksget_task_logRead it →
Codexon the RenderFlow farm
“Requeue every frame that failed overnight.”Finds 14 failed tasks across three jobs, shows them, and requeues them after your yes.
find_taskscontrol_tasksRead it →
Claudeon the RenderFlow farm
“Push the hero shot to the top of the queue.”Raises the job above everything else waiting, without touching what it has rendered.
find_jobsupdate_jobsRead it →
Cursoron the RenderFlow farm
“Submit this Blender scene, frames 1 to 240, to the GPU pool.”Checks the path every machine can reach, then submits the job once you confirm.
farm_overviewsubmit_jobRead it →
Claudeon the RenderFlow farm
“Did any frames of Hero_CPU_v07 come out black?”Looks at the finished frames as images and finds the one that came out black.
find_tasksget_task_previewRead it →
LM Studioon the RenderFlow farm
“How busy is the farm right now?”One call for the whole farm: what is rendering, what is waiting and which machines are free.
farm_overviewRead it →
Claudeon the RenderFlow farm
“Put the workstations on the farm every night at 7 PM.”Plans a schedule event for the workstations and shows you what it will do before saving it.
list_schedulecontrol_scheduleRead it →
VS Codeon the RenderFlow farm
“Which machines keep failing?”Reads the farm-wide error log, folds the repeats and points at the machine to look at.
find_errorsfind_nodesRead it →
Codexon the RenderFlow farm
“This V-Ray error keeps coming back. Can we catch it?”Writes an error catalog entry for the next time, and leaves it off until you switch it on.
error_catalogRead it →
Claudeon the RenderFlow farm
“What happened on the farm while I was away?”Reads the event log and sums up who started, stopped and moved what.
find_eventsRead it →
Claudeon the RenderFlow farm
“Stop Lobby_v12, the client changed the camera.”Says what stopping does to the frames under way, then stops the job after your yes.
find_jobscontrol_jobsRead it →
Codexon the RenderFlow farm
“Clear last month’s finished jobs out of the list.”Archives them rather than deleting them, so any of them can come back.
find_jobscontrol_jobsRead it →

Admin-controlled

You decide what it can do

AI clients are off on a new farm. An admin turns them on, and decides who else may connect their own assistant.

Off until you turn it on

Off until you turn it on

Settings > Security decides whether the farm answers AI clients at all.

It acts as that computer

It acts as that computer

On a machine that runs RenderFlow, the assistant can do what the RenderFlow app there can do. No key, no account.

Keys with scopes

Keys with scopes

A laptop without RenderFlow connects with an API key and does only what the key allows.

Nothing deletes

Nothing deletes

It can start, stop, requeue and reschedule, but nothing it does deletes anything.

Talk to us about access

There’s more to RenderFlow

The same farm renders the rest of your pipeline

Cinema 4D, 3ds Max, Blender, Unreal Engine, Nuke and every other application your team works in.

3ds MaxAfter EffectsArnoldAutographBlenderCinema 4DComfyUIDaVinci ResolveMedia EncoderFFmpegFusionHarmonyHoudiniKeyShotMayaNukePremiere ProPythonRedshiftRenderManRhinoSketchUpUnreal EngineVantageV-RayVRED

Your whole farm in one view

Ten machines or thousands, every job and machine in one place, with statistics and a scheduler.

Farm management →

Built for the whole team

Artists, developers, IT admins and supervisors each get the view they need.

For every role →

More power when the farm is full

Cloud machines from your own Azure or AWS account, started only when you need them.

Hybrid cloud →

Your AI render wrangler

Ask Claude or Codex why a frame failed, retry it, or schedule the night machines. It asks first.

AI integration →

Explore RenderFlow

Explore RenderFlow

Put an AI wrangler
on your farm

Try RenderFlow

Try RenderFlow

Book a demo

Book a demo

AI wrangler questions

Which AI clients work?

Claude Desktop, Claude Code, Codex, Cursor, VS Code, LM Studio and any client that speaks MCP over streamable HTTP.

Can ChatGPT connect?

Not to a farm on your network. ChatGPT reaches MCP servers from its own servers on the internet, and so do Claude's Connectors. Use Claude Desktop or Claude Code instead.

Does it learn our errors?

When it finds a cause your error catalog doesn't know, it offers a new entry and a sanity check that catches the problem before the next submit. You see both first, and a new entry starts switched off.

Can it change things without asking?

No. It shows you what it will do and waits for your yes, and nothing it can do deletes anything.

Explore all FAQs

Still have a question? Talk to us →