Local-first · Agent Runtime
AI-driven stable workflows.
Avoid black-box AI results. Build stable workflows anyone can design, with Help Thing refining ideas and requirements.
Local executionHuman approval for risky actionsNode-level recovery
iAgent · stable workflow runtimerunning locally
User request
Turn this recurring job into a stable workflow.
01Describe jobtrigger.intent
02Resolve ambiguityllm.semantic_step
03Run tooltool.deterministic
04Human approvalhitl.approval
05Failure recoveryruntime.recovery
Node stateEvidence saved
01Describe jobdone
02Resolve ambiguityverifying
03Run toolaudited
04Human approvalwaiting
05Failure recoveryready
Why iAgent
For AI to land reliably, the runtime needs control.
One-shot answers do not become durable automation. iAgent separates flow, models, tools, approval, and recovery so every step is visible, pausable, and traceable.
01 · control / runtime
Code controls certainty
State, permissions, tool execution, persistence, logs, verification, and recovery are owned by code, not model improvisation.
02 · model / ambiguity
Models handle ambiguity
The LLM handles understanding, extraction, explanation, and repair suggestions, without owning the whole control flow.
03 · approval / human
People approve risk
Write, overwrite, send, delete, and high-impact actions pause until a person reviews the input, output, and risk.
04 · recovery / audit
Failures can recover
Each node has state, inputs, outputs, local memory, and records. Failures locate the node, suggest repairs, and rerun from the right point.
How it works
From one sentence to one stable workflow.
Draft the flow, review and test it, then keep approval, verification, and node-level reruns inside one runtime loop.
01
Describe the job
The user explains the outcome. iAgent turns the goal, inputs, and success criteria into a workflow draft.
Turn this recurring job into a stable workflow.
I will draft:
· Inputs and success criteria
· Model, tool, and approval nodes
· Inputs and success criteria
· Model, tool, and approval nodes
Workflow draft generated
02
Review and test
The user sees who owns each step: code, model, or human. A test run records node state before release.
03
Approve, verify, rerun
Risky actions pause for approval. Failures point to a node, show evidence, and rerun after review.
NodeOwnerState
01Input sourceDone
02Model stepNeeds check
03Tool stepAudited
04Human approvalWaiting
05VerificationNot started
3 done · 1 pendingContinue run
Pricing
Pricing with one clear currency per page.
The current public page keeps Free, Plus, Pro, and Pro+. The English page shows USD only; final numbers and limits follow the backend public plan source.
Free · trial / invite-only
$0software
Good for validating workflow fit first
- Local workflow editing and review
- Access to launch-scope flows
- Desktop update alerts
- Final numbers follow the public plan source
Good for light trials
Plus · light paid usage
$9from
Good for solo users starting real workflows
- Full desktop app
- Local-first workflows
- Human review on critical steps
- Final numbers follow the public plan source
Paid tiers stay local-first
Launch focus
Pro · launch recommended
$19from
Best fit for most regular workflows
- All launch nodes and approval flows
- Higher usage allowance
- Sync capability as defined in launch scope
- Final numbers follow the public plan source
Backend public plans remain the source of truth
Pro+ · heavy usage
$99from
Built for frequent calls and heavier workflows
- Highest launch usage allowance
- Better fit for power usage
- Priority support
- Final numbers follow the public plan source
Built for heavier workflow usage
Runtime principles
iAgent is valuable because the work becomes controllable.
01
Repeatable
The same job does not need a fresh prompt every time. Once the flow is fixed, new inputs still follow clear control logic.
02
Reviewable
The model only handles ambiguous judgment. Critical steps pause for people, and the decision becomes part of the record.
03
Recoverable
Failure does not mean starting over. Locate the node, inspect evidence, apply a reviewed fix, and rerun from there.
FAQ
Common questions.
A chat tool returns an answer. iAgent gives you a reviewable, testable, rerunnable workflow where every step has state, inputs, outputs, and records.
No. The model handles ambiguity such as understanding, extraction, explanation, and repair suggestions. The runtime and the user control flow, permissions, tools, state, and risky actions.
Some actions create real consequences. iAgent pauses at risky nodes so you can inspect what the step will do before approving or rejecting it.
Failure lands on a specific node. You can inspect input, output, error, and impact, ask AI for a repair suggestion, review it, and rerun from that node.
The default is local-first processing. Only model-backed steps send the smallest useful context needed for that step.
The site does not present broken installer links as downloads. You can check desktop status on the download page or sign in to try Web Chat first.
Yes. Web Chat is useful for lightweight questions and temporary file analysis. Full local-first workflows remain centered on the desktop app.