Plan gate, metering and code check
Professional and Enterprise; one AI unit per build; generated code is checked for forbidden calls before deploy.
Gate does not cover save, knowledge, test runs or logs
AI agents
AI agents you build, test and deploy, then call from any flow
Describe an agent or start from one of 8 templates; edit its code, input and output schemas, packages and environment in the builder; run payload tests; let the build loop generate, deploy a test service, run your cases and fix failures; give it knowledge from PDFs, pages, text and your own workspace data; deploy it as its own service with logs; call it from flows or on a schedule. Professional and Enterprise.
Free plan. No credit card required.
Aarushiya Healthtech is a fictional company used for illustration.
- 8starter agents
- One builderfor code, schemas, packages and env
- Build loopgenerate, test, fix
- Knowledgefrom files, pages and your data
- Your modelstrained on your data, deployed as agents
- Any flowruns an agent as a step
Templates
Start from a working agent
My Agents and Templates with search and status filter. Eight starter agents: survey insight, email intelligence, lead qualifier, campaign content generator, data transformer, customer experience analyzer, webhook intelligence, RAG knowledge agent. Use a template, then change anything.
- My Agents and Templates with search and a status filter.
- Survey insight, email intelligence, lead qualifier, campaign content generator, data transformer, customer experience analyzer, webhook intelligence, RAG knowledge agent.
Needs a paid plan: Professional · EnterpriseSee plans
Everything in Templates2
- Agent catalogProfessional · EnterpriseMy Agents and Templates with search and a status filter.
- 8 starter agentsProfessional · EnterpriseSurvey insight, email intelligence, lead qualifier, campaign content generator, data transformer, customer experience analyzer, webhook intelligence, RAG knowledge agent.Each template needs the workspace's own OPENAI_API_KEY
Builder
A real editor, not a prompt box
Chat with the assistant on the left; on the right a code editor over index.js and package.json, an input schema and an output schema, environment variables with masked values and auto-detection of the ones your code reads, and the knowledge tab. Autosave, a version counter on every save and a code hash on every deploy.
- A chat panel beside the editor drafts and changes the agent from plain language.
- Edit index.js in a full code editor with syntax highlighting.
- Define the input payload and the output shape; inputs drive flow mapping and test payloads, outputs become flow fields.
- Edit package.json; missing packages your code requires are added at build with their npm versions.
- Masked values, variables your code reads detected automatically, names normalised.
Needs a paid plan: Professional · EnterpriseSee plans
Everything in Builder5
- Describe the agent in chatProfessional · EnterpriseA chat panel beside the editor drafts and changes the agent from plain language.
- Code editorProfessional · EnterpriseEdit index.js in a full code editor with syntax highlighting.Node 18 runtime, fixed
- Input and output schemasProfessional · EnterpriseDefine the input payload and the output shape; inputs drive flow mapping and test payloads, outputs become flow fields.Output is not validated at run time
- Package definitionProfessional · EnterpriseEdit package.json; missing packages your code requires are added at build with their npm versions.Dependency allow-list is advisory only
- Environment variablesProfessional · EnterpriseMasked values, variables your code reads detected automatically, names normalised.Values are stored in plain text
Aarushiya Healthtech is a fictional company used for illustration.
Build loop
Build, test and fix in one loop
Press Build: the agent is generated from your description and schemas, deployed to a test service, your test cases run against it, and failures go back to the model, up to three rounds. Missing environment variables are flagged before anything deploys. One AI unit per build.
- Generate the agent, deploy a test service, run the test cases, feed failures back, up to three rounds.
Needs a paid plan: Professional · EnterpriseSee plans
Everything in Build loop1
- Build, test and fix loopProfessional · EnterpriseGenerate the agent, deploy a test service, run the test cases, feed failures back, up to three rounds.1 AI unit per build request; each round runs Cloud Build (minutes)
Aarushiya Healthtech is a fictional company used for illustration.
Testing
Test with real payloads
Several test cases per agent, each an input JSON and an optional expected output; auto-generate a payload from the input schema; run one or all; pass or fail per key; cases are saved with the agent.
- Several cases per agent, input JSON plus optional expected output; auto-generate from the schema; run one or all; cases saved with the agent.
Needs a paid plan: Professional · EnterpriseSee plans
Everything in Testing1
- Test cases with payloadsProfessional · EnterpriseSeveral cases per agent, input JSON plus optional expected output; auto-generate from the schema; run one or all; cases saved with the agent.Pass/fail is exact equality per key
Knowledge
Give it your knowledge
Add URLs, PDFs up to 10 MB or pasted text: chunked, embedded and ranked at answer time. Add internal sources too: a survey, a contact list, an import or messages, queried live with filters mapped from the agent's inputs.
- Add URLs, PDFs up to 10 MB or pasted text; chunked, embedded and ranked when the agent answers.
- Point the agent at a survey, a contact list, an import or messages, queried live with filters mapped from its inputs.
Needs a paid plan: Professional · EnterpriseSee plans
Everything in Knowledge2
- Knowledge from pages, PDFs and textProfessional · EnterpriseAdd URLs, PDFs up to 10 MB or pasted text; chunked, embedded and ranked when the agent answers.
- Knowledge from your workspace dataProfessional · EnterprisePoint the agent at a survey, a contact list, an import or messages, queried live with filters mapped from its inputs.
Deploy and logs
Deployed as its own service, with logs
Deploy creates a container and a service of its own for the agent, in our cloud; undeploy stops it; the logs console shows its output with a severity filter; test deployments clean themselves up after seven days.
- One click builds a container and runs the agent as its own service in our cloud; redeploy any time.
- Stop the service and mark the agent undeployed; test deployments clean themselves up after 7 days.
- Read the agent's production or test logs with severity, since and limit filters.
- Every save bumps the version; every deploy stamps a code hash; autosave 10 s after a change.
Needs a paid plan: Professional · EnterpriseSee plans
Everything in Deploy and logs4
- Deploy as its own serviceProfessional · EnterpriseOne click builds a container and runs the agent as its own service in our cloud; redeploy any time.Cloud Run defaults for scaling (no min/max settings)
- UndeployProfessional · EnterpriseStop the service and mark the agent undeployed; test deployments clean themselves up after 7 days.
- Logs consoleProfessional · EnterpriseRead the agent's production or test logs with severity, since and limit filters.
- Version counter and code hashProfessional · EnterpriseEvery save bumps the version; every deploy stamps a code hash; autosave 10 s after a change.No version history, diff or rollback
Aarushiya Healthtech is a fictional company used for illustration.
In flows
A step in any flow
Add 'Run an AI agent' to a flow: pick a deployed agent, map each input to a field of the record, and every output becomes a field for the steps after it. Agents can also be the source of a scheduled flow. Two flow templates ship with an agent step. You cannot delete an agent a flow still uses.
- Pick a deployed agent, map each input to a field of the record, use every output in later steps and conditions.
- A schedule can call an agent and run the flow on what it returns.
- AI Agent Response and Feedback Loop with AI Analysis ship with an agent step.
- Deleting an agent that flows still reference is blocked, with the list of dependents.
AarushiyaThe warranty assistant reads the policy PDF and answers the first WhatsApp question on every claim.
Needs a paid plan: Professional · EnterpriseSee plans
Everything in In flows4
- Run an AI agent in a flowProfessional · EnterprisePick a deployed agent, map each input to a field of the record, use every output in later steps and conditions.One record per run; outputs not schema-validated
- Agents as a scheduled sourceProfessional · EnterpriseA schedule can call an agent and run the flow on what it returns.
- Flow templates with an agent stepProfessional · EnterpriseAI Agent Response and Feedback Loop with AI Analysis ship with an agent step.
- Cannot delete an agent a flow usesProfessional · EnterpriseDeleting an agent that flows still reference is blocked, with the list of dependents.
Aarushiya Healthtech is a fictional company used for illustration.
Questions buyers ask
Where does an agent run?
One click builds a container and runs the agent as its own service in our cloud; redeploy any time. Node 18 runtime. Test deployments clean themselves up after 7 days.
What can it read?
URLs, PDFs up to 10 MB or pasted text, chunked and packaged with the agent. Point it at a survey, a contact list, an import or messages, queried live with filters mapped from its inputs.
Which plan?
Professional and Enterprise; one AI unit per build.
Do I need my own OpenAI key?
Yes. Each template needs the workspace's own OPENAI_API_KEY; the templates read it from the agent's environment.
Can I start from a ready-made agent?
Eight starter agents: survey insight, email intelligence, lead qualifier, campaign content generator, data transformer, customer experience analyzer, webhook intelligence, RAG knowledge agent. Use a template, then change anything.
What does Build do?
Generate the agent, deploy a test service, run the test cases, feed failures back, up to three rounds. Missing environment variables are flagged before anything deploys. Each round runs Cloud Build (minutes).
How do I test an agent?
Several test cases per agent, each an input JSON and an optional expected output; auto-generate a payload from the input schema; run one or all; pass or fail per key; cases are saved with the agent.
How does a flow use an agent?
Add 'Run an AI agent' to a flow: pick a deployed agent, map each input to a field of the record, and every output becomes a field for the steps after it. One record per run; outputs are not schema-validated.
What happens if I delete an agent a flow uses?
Before deleting, the platform lists every flow that uses the agent; if any of them are active, it warns you and asks you to deactivate them first or explicitly choose Delete Anyway. Inactive flows are only shown as a note.
Can I see what a deployed agent does?
Read the agent's test or production logs with severity and time-range filters; the console shows up to 200 entries.
Start from a working agent
Use a template, then change anything.
Free plan. No credit card required.