FAQ

General

What is Fruxon?

Fruxon is a platform for building, shipping, and operating AI agents — visual design, any LLM, real integrations, production-grade observability and controls. See the welcome page for the elevator version.

Who is Fruxon for?

Teams building real AI workflows that need to interact with their actual systems — not just LLM playgrounds. Engineers, ops teams, support teams, and product teams use Fruxon to ship agents into production with the safety nets they'd want from any serious infrastructure.

Do I need to know how to code?

No, for most workflows. Studio is fully visual — you wire steps with placeholders, configure prompts, and test interactively. You can build complex agents without writing code.

You'll want code if you're integrating Fruxon into your product (use the API or the Python SDK), or if you need a Code Executor step to run custom logic.

What models does Fruxon support?

OpenAI, Anthropic (Claude), Google (Gemini), AWS Bedrock, DeepSeek, xAI (Grok), HuggingFace, and VoyageAI. You bring your own keys; Fruxon never resells inference. The exact model catalog changes as providers ship new models, so rather than list specific names here, see AI Providers for the current lineup.

Does Fruxon train models on my data?

No. Your prompts and outputs flow between Fruxon's runtime and your chosen provider; nothing is used to train Fruxon's or anyone else's models. Provider data policies apply for inference (e.g., OpenAI's enterprise default of not training on API inputs).

Pricing & limits

How does pricing work?

Plans cover platform features (seats, retention, eval runs, etc.). LLM costs are passed through — you bring your own provider keys, so token spend goes directly to OpenAI / Anthropic / etc. Fruxon shows you the per-run cost in every trace using the public model pricing table.

Is there a free tier?

Yes. The free tier is enough to build a few agents, test them, and run them at modest volume. See pricing on fruxon.com.

What are the rate limits?

API rate limits vary by plan and are documented in Settings → Usage. Connector throughput is generally bounded by the channel itself (e.g., Slack's per-channel limits). Provider rate limits are separate and apply to your direct usage of OpenAI / Anthropic / etc.

Can I set spending caps?

Yes — per-agent monthly budgets with alert thresholds and hard caps. See Cost & Budgets.

Building agents

How are agents different from prompt chains?

An agent has typed inputs, multiple steps with placeholder-driven dataflow, real tools (APIs, sub-agents, code), persistent memory, attached knowledge, and revision history. A prompt chain is one shape an agent can take. Agents go beyond chains.

Can I have an agent call another agent?

Yes — sub-agents. Any agent can be called as a tool from any other agent in your organization. See Sub-agents.

Can I import existing prompts?

Paste them into Studio. There's no migration tool from other platforms today, but the visual builder makes it fast to recreate flows.

Can I run code inside an agent?

Yes — the Code Executor tool runs Python in a secure sandbox. Use it for transforms, charts, calculations, custom parsing.

How do I make my agent remember things?

Two options:

• Memory — persistent key-value storage scoped to the agent, written and read via tools. (Memory)
• Sessions — multi-turn conversation state, automatically managed and summarized. (Sessions)

Use memory for facts ("the user's preferred timezone is PST"); use sessions for the recent conversational context.

Connecting & integrating

What integrations are available?

45+ across CRM (Salesforce, HubSpot, Zoho), comms (Slack, Discord, Intercom), project tools (Asana, Monday, Linear, Jira, ClickUp), databases (PostgreSQL, MySQL, MongoDB, ArangoDB, Redis, Supabase), search (Tavily, BrightData, Apollo), docs (Confluence, Notion, GitHub, Google Drive), and more. See Integrations.

Can I connect a custom REST API?

Yes — import an OpenAPI/Swagger spec, or configure individual endpoints by hand. The agent gets the endpoints as callable tools.

What channels can I deliver agents on?

Slack, Microsoft Teams, Telegram, WhatsApp (via Twilio or 360Dialog), Discord, plus the REST API and the embeddable web chat widget. See Agent Network.

Can the same agent serve multiple channels at once?

Yes. Bind multiple participants, participant channels, and triggers to one agent; each can deliver the same deployed revision through its own channel or event source.

Production

How do I roll back a bad deploy?

Open Versioning, find the previous good revision, click Deploy. The switch is atomic. (Versioning)

How do I test a candidate revision before deploying?

Run it interactively in the Studio test runner, or score it with an evaluation against your golden dataset and compare it to the deployed baseline. The execute API always targets the deployed revision, so promote (deploy) once you're confident — and roll back instantly if needed.

How do I monitor production?

Observability shows every run with full traces. Conversations shows end-user sessions. Set alerts on errors and budgets.

Does Fruxon have an SLA?

Plan-dependent. Enterprise plans include uptime SLAs and dedicated support. Contact your account team.

Security & compliance

Where is my data stored?

In Fruxon's managed cloud (region depends on your plan). All data is tenant-isolated, encrypted at rest, and accessed only by your authenticated members. Enterprise teams can choose a dedicated single-tenant deployment with region pinning, or run Fruxon inside their own cloud — see Hosting & Data Residency.

Is Fruxon SOC 2 / GDPR / HIPAA compliant?

Fruxon has completed a SOC 2 Type 1 examination (report available under NDA) and is verified against CASA Tier 2 (Google's Cloud Application Security Assessment). For GDPR, a standard Data Processing Addendum (DPA) is available, along with the current subprocessor list and a pre-filled CAIQ/SIG security questionnaire on request. The baseline includes TLS 1.2+ in transit, AES-256 at rest, RBAC, MFA, 24/7 centralized logging, audit trails, and vulnerability scanning.

Fruxon is not HIPAA-attested and does not claim HIPAA compliance. Secrets and credentials can be tagged with a PHI sensitivity tier that enforces stricter, fail-closed audit handling (see Secrets → Sensitivity tiers), but that tier is an internal control — not a HIPAA attestation, so don't treat it as one.

Reports, the DPA, and the subprocessor list are available at fruxon.com/trust, or email security@fruxon.com.

How do I handle PII?

Redact in pre-processing steps, or design your agent to never pass PII to the LLM. Fruxon will retain whatever the trace captures, so be intentional. See Security.

Can I delete a user's data on request?

Yes. Filter conversations by end-user, select all, delete. The action is logged.

Team & access

How do I invite teammates?

Settings → Team Management → Invite Member. Pick an org role — Admin or Member. Per-agent access (Viewer / Editor / Admin) is granted separately on each agent. They get an email link.

Can I have separate dev and prod environments?

Use separate organizations. A user can belong to many. See Team patterns.

Can different team members own different agents?

Not yet — all agents in an organization are shared by all members. Use multiple organizations if you need ownership separation.

Developer

Is there a CLI?

Yes — pip install fruxon ships the fruxon command. It covers sign-in, agent discovery, one-shot execution, an interactive chat REPL, draft-flow testing, trace inspection, and doctor diagnostics. See the CLI guide.

Is there a TypeScript / JS SDK?

Not yet. Use fetch against the REST API directly. The OpenAPI spec is published and importable into your tooling.

Can I receive webhooks for agent events?

Webhook delivery for agent events (run completed, conversation started, etc.) is on the roadmap. Today, poll Execution History via the API or use the Python SDK's streaming endpoints.

Can I self-host Fruxon?

Fruxon runs as managed SaaS on Google Cloud today, with per-tenant isolation. A self-hosted / customer-VPC deployment — Fruxon-packaged and customer-operated — is offered as an enterprise engagement rather than a turnkey self-serve product; the deployment architecture is scoped per engagement. See Hosting & Data Residency for the deployment models, key custody, and residency options, or contact sales to start the conversation.

Still stuck?

• Troubleshooting — debugging recipes for common issues
• Glossary — definitions of every term in these docs
• support@fruxon.com — include your organization slug and an execution ID for fastest triage