Checking your session...
Setup checklist Review the saved profile, knowledge and test evidence.
Loading profile...
Check the linked sources and model configuration.
Use the saved profile and review fictional test evidence before a pilot.
Other agent jobs and builders
Registered jobs
Off stops a job. Suggest runs only its defined work under current permissions. Each job retains its own controls and history.
Loading...
Suggest an agent improvement
Propose fictional example wording for an administrator to review. Suggestions do not change guidance. Quasar AI Work and access to the agent's current module, private project and sources are required.
Support reply agent
Give the Inbox reply job its role and guidance. A person reviews and sends every reply.Uses the customer's conversation and permitted read lookups. Link a private project to use its instructions, selected issued policies and reference manuals under the assigned person's current access. Rules cannot grant permissions or change financial and warranty decisions.
Loading saved Support profile...
5. Test, review and improve
Guidance history
Browse saved wording, compare versions and load earlier guidance into the editor. Loading does not save it. Review the result and use Save profile to retain a new version.
Save a profile to create its first guidance version.
Test agent
Try a fictional customer conversation against the saved profile and its private project. This manual test sends the sample and permitted guidance to the saved provider and model and incurs API usage. It creates no customer activity, quotes, invoices or sent messages. Choose guidance only, or rehearse a conversation with fictional customer, order and catalogue tools. Real stock, pricing and delivery are never verified here.
Guidance only. No business lookups.
Load fictional test pack
Prepare up to 20 fictional cases from a JSON file or pasted JSON. Loading stays on this page and does not save settings or start a model request. Review the file first; use no real customer details, private correspondence or credentials. Cases are not automatically anonymised.
Paste JSON instead
Paste the contents of a fictional JSON pack, up to 200,000 UTF-8 bytes. This uses the same checks as a file import and stays on this page. No profile is saved and no model request starts.
No pack loaded. Local preparation is available before a profile is saved. Run still requires the existing saved profile, project, source and provider checks.
Choose a case, review its expected behaviour, then use it as the sample. Only Run test draft starts a paid request. Packs are not retained after leaving or reloading this page; completed runs and reviews use the existing private test history.
Use fictional details, up to 6,000 characters. Never paste credentials. Save tone, evidence checks and handover edits above before running a fresh comparison.
One test at a time for the team; two per person in ten minutes; ten per person and up to ten for the team per Brisbane day. Every attempt reserves 100,000 budget tokens, including failures. Reported model usage is shown separately.
My recent tests
Review and improve
Review a completed test above, save useful cases and rerun them after guidance changes. Approval saves a new profile revision; an active Suggest job uses it for future drafts.
Saving retains a rules revision and leaves the switch unchanged. When Inbox email replies is on Suggest, eligible assigned inbound conversations use these rules for draft preparation. Load a file to edit the form, then save to apply it.
Build an internal assistant
Choose a role, personality, task rules and private knowledge in Quasar AI Chat. Authorised creators and editors manage the guidance; administrators appoint who can use it. Current staff permissions control every tool.
Open assistant builderSales reply agent
Give the Sales reply job its role and guidance. A person reviews and sends every reply.Uses the customer's conversation and permitted read lookups. Link a private project to use its instructions, selected issued policies and reference manuals under the assigned person's current access. Rules cannot grant permissions or change financial and warranty decisions.
Loading saved Sales profile...
5. Test, review and improve
Guidance history
Browse saved wording, compare versions and load earlier guidance into the editor. Loading does not save it. Review the result and use Save profile to retain a new version.
Save a profile to create its first guidance version.
Test agent
Try a fictional customer conversation against the saved profile and its private project. This manual test sends the sample and permitted guidance to the saved provider and model and incurs API usage. It creates no customer activity, quotes, invoices or sent messages. Choose guidance only, or rehearse a conversation with fictional customer, order and catalogue tools. Real stock, pricing and delivery are never verified here.
Guidance only. No business lookups.
Load fictional test pack
Prepare up to 20 fictional cases from a JSON file or pasted JSON. Loading stays on this page and does not save settings or start a model request. Review the file first; use no real customer details, private correspondence or credentials. Cases are not automatically anonymised.
Paste JSON instead
Paste the contents of a fictional JSON pack, up to 200,000 UTF-8 bytes. This uses the same checks as a file import and stays on this page. No profile is saved and no model request starts.
No pack loaded. Local preparation is available before a profile is saved. Run still requires the existing saved profile, project, source and provider checks.
Choose a case, review its expected behaviour, then use it as the sample. Only Run test draft starts a paid request. Packs are not retained after leaving or reloading this page; completed runs and reviews use the existing private test history.
Use fictional details, up to 6,000 characters. Never paste credentials. Save tone, evidence checks and handover edits above before running a fresh comparison.
One test at a time for the team; two per person in ten minutes; ten per person and up to ten for the team per Brisbane day. Every attempt reserves 100,000 budget tokens, including failures. Reported model usage is shown separately.
My recent tests
Review and improve
Review a completed test above, save useful cases and rerun them after guidance changes. Approval saves a new profile revision; an active Suggest job uses it for future drafts.
Saving retains a rules revision and leaves the switch unchanged. When Sales enquiry replies is on Suggest, eligible assigned inbound conversations use these rules for draft preparation. Load a file to edit the form, then save to apply it.
Production planning agent
Review build quantities before committing materials, staff or production orders.Production planning has its own rules and proposal controls. Support and Sales replies have separate profiles above. Other jobs keep their current controls.
Role, rules and AI provider
Write instructions in plain English or load a Markdown (.md) file. These describe the job; they cannot grant new permissions. API keys belong in the server's secret settings.
Loading profile...
Saved proposals
These are dated reviews. Refresh live demand before deciding what to build. No production order or staff booking is created here.
Voice preview, on hold
Voice rollout is on hold. These existing internal preview tools remain available for a separately reviewed setup. Opening this section does not start a session or enable voice.
Voice preview
Build a customer-safe voice profile and test it internally. Quasar owns the answers. Telephone routing and customer rollout need a separate reviewed setup.
Customer communication controls
Reviewed delivery sends only the exact email a person has approved in Quasar. Sales follow-ups prepare drafts for review. Both start off. Brain switches, current access, private knowledge, customer stop controls and human outcome decisions still apply.
Support
Sales
Use AI outreach in Conversations to enrol an opportunity, review an exact email, stop customer outreach or prepare an invoice handover. Delivery is checked every five minutes and limited to 20 attempts per team and 10 per approving person per Brisbane day. Financial documents and sensitive outcomes use the existing direct native workflows. Changing saved delivery controls invalidates older approvals. Switching a Brain job off stops its next execution check; an email already in flight can still complete.
Custom task agents
Build a private knowledge brief or an assigned Support or Sales review. Tasks use their named owner's current access and the selected private project's rules and issued sources. They save internal reports for review. They cannot send messages, change business records or approve outcomes.
The Custom task engine must be on Suggest and the task must be On before Run now or a schedule can execute. Switch off takes effect at the next execution check. A provider request already in progress may still incur usage; its result is withheld if the task or access changes.