GUARANTEE 01
Same guardrails
Eight rail types fire on every agent regardless of build path. PII redaction, content safety, jailbreak detection, grounding. Same Colang rules, same NeMo NIMs.
Ready to get started?
Deploy sovereign AI on your infrastructure - in weeks, not months.
Developers · Build an agent · No-code to BYOA
Wizard, conversational, workflow canvas, SDK, or BYOA - every path produces the same agent package on the same runtime.
Pick your level. Keep your governance.
Most platforms force a single build experience. Pick the SaaS console or the SDK; you can't have both. Pick LangChain or CrewAI; you can't switch. Katonic accepts five inputs and produces one output: the NeMo Agent Toolkit workflow package. Same input contract, same telemetry, same policy surface.
What every agent inherits, regardless of path
Eight rail types fire on every agent regardless of build path. PII redaction, content safety, jailbreak detection, grounding. Same Colang rules, same NeMo NIMs.
Tool calls pass through the governance proxy. Six policy actions, five risk levels, HITL approvals when required. Identical policy surface.
Auto-instrumentation traces every tool call, LLM call, and decision via OpenTelemetry. Streams to your existing observability tools. No opt-in.
PATH 01
For
Business analystCode
NoneHow
7-step wizardSample
PATH 02
For
AnyoneCode
NoneHow
ConversationalSample
PATH 03
For
Pipeline architectCode
Config onlyHow
DAG canvasSample
PATH 04
For
Python developerCode
Python + YAMLHow
Embedded IDESample
agent: hr-policymodel: balancedtools: [...]PATH 05
For
Existing agent teamCode
ExistingHow
Upload + scaffoldSample
This is what your team sees in Studio. Five build surfaces, every one of them publishing the same NAT workflow package. Toggle between them; the chrome you'd see at app.your-org.katonic.ai/studio is what's rendered below.
HR Policy Assistant
draft · v0.4 · auto-saved 12s ago
Step 4 · Knowledge
50+ connectors · 13 native, 30+ via Onyx adapter. Index now or schedule incrementals.
Employee Handbook 2025
SharePoint · /HR-Hub · 284 chunks · indexed 2h ago
Benefits & Compensation
SharePoint · /Benefits · 147 chunks · indexed 1d ago
Leave Policy v2.3
Confluence · HR Space · 62 chunks · indexed 5h ago
Test chat · live
📄 source · Employee Handbook 2025, p.47
/studio/builder, /studio/ai-builder, /studio/workflow, /studio/code-builder, and /studio/byoa in your sandbox today. Same NAT workflow package out of all five.A form-driven wizard for the business analyst. Pick a template from the marketplace; the seven steps fill in around it. Override what you need, ship what you have. The output is a versioned agent record that publishes the same package format every other path produces.
240-plus MCP servers in the catalog. Search, filter by category, expand to choose specific tools.
sharepoint-hr
8 tools · Knowledge
sharepoint-policies
5 tools · Knowledge
sharepoint-finance
6 tools · Knowledge
Live config
HR Policy Assistant
4 steps remaining
80-plus pre-built agents. 150-plus templates. Pick the one closest to what you need, override the prompt, swap the knowledge source, ship.
A chat interface that turns a natural-language description into a complete agent configuration. Each turn refines the spec. The config file updates live. When you're satisfied, publish.
AI Builder · session #28f4
● ACTIVEI need an HR assistant that searches SharePoint and answers policy questions about leave, benefits, and compensation.
Got it. I'm drafting HR Policy Assistant. Three quick questions:
balanced. Override?Yes citations. Escalate compensation. Keep balanced.
Configuration ready. Reviewing the config file on the right▋
agent: name: "hr-policy-assistant" owner: "@hr-team" model: tier: "balanced" tools: - server: "sharepoint-hr" knowledge: - "Employee Handbook 2025" cite_sources: true escalation: - topic: "compensation" to: "hitl" guardrails: "strict"
Other platforms offer "describe and deploy" that auto-optimises a fixed pipeline. The AI Builder is iterative: each turn refines the spec, and the underlying config is editable at any point. Drop into the Code Builder mid-conversation to take over.
A full workflow canvas for multi-step builds. 13 typed node types, 4 categories, container nodes that group sub-workflows. Cycle detection and validation surface errors per-node before publish. Output: the same standard agent config.
An embedded IDE with syntax highlighting, file management, terminal, one-click test/deploy. The developer maintains a config file for declarative resource bindings and agent.py for the actual logic. The Katonic SDK resolves human-friendly names to platform endpoints behind the scenes.
agent: name: "hr-policy-assistant" model: tier: "balanced" tools: - server: "sharepoint-hr" tools: [search_documents, get_policy] - server: "hris-system" tools: [get_balance] knowledge: - "Employee Handbook 2025" guardrails: "strict"
Terminal · resolve config
→ Binding sharepoint-hr to MCP layer... ok
→ Binding hris-system to MCP layer... ok
→ Binding "Employee Handbook 2025" to knowledge engine... ok
→ Binding balanced tier to AI gateway... ok
✓ Agent package built · 4 resources bound · 0 errors
$ ▋
Resolver, not framework. tools.mcp() returns a binding; gateway.model() returns a binding; knowledge.source() returns a binding. The runtime does the rest. SDK reference →
Teams already running LangChain, CrewAI, or LangGraph agents don't need to rewrite. Drop the .py or .zip in. The BYOA service auto-detects the framework, extracts tools and model references, scaffolds the platform package around the existing code, publishes. The agent gains every platform capability on import.
refund-agent.zip
4.2 MB · uploaded 2s ago · sha256:9a31...
CrewAI
confidence 0.97
3 tools
1 model
Agent package
generating files...
Publish
queued
Eight rail types · 6 policy actions · 5 risk levels · HITL approvals · 7 evaluators · OpenTelemetry traces · audit log · model lineage tracked. Without rewriting a line of your CrewAI code.
Whether you're in the Guided wizard, the AI Builder chat, the DAG canvas, the IDE, or BYOA, four tools follow you across every path: prompt management, optimisation, component library, marketplace starters.
The matrix · how tools connect to paths
Auto-versioned prompt CRUD. Side-by-side model comparison. Batch evaluation against datasets. AI-powered prompt improvement with rationale.
Automated prompt refinement via Optuna GridSearch. Jointly optimises prompt wording, model selection, and temperature for the best quality-cost tradeoff.
20 pre-built UI components across 6 categories. Plus the 4-step Component Builder for adding your own with AI-assisted schema.
80-plus pre-built agents organised by department and industry. 150-plus templates. 6 framework starter kits.
Building the agent is one part. Owning it through dev, test, prod, and the inevitable rollback is the other. Every agent gets immutable version snapshots, hot-reloaded publishes, eval-gated promotion, one-click rollback, deep clone, and portable JSON export.
Lifecycle stages · explained
Version
Every save freezes config into an immutable AgentVersion. Snapshot includes prompt, tools, knowledge, guardrail profile.
Publish
An event fires. The agent runtime hot-reloads without restart. Zero downtime.
Promote
dev → test → prod with eval gate enforcement. Below threshold? Blocked. Promotion copies config; target env runs its own ingestion.
Rollback
Re-point the active version to any previous one. One click. Hot-reloaded the same way as a forward publish.
Clone
Deep copy across environments or teams. Useful for forking a prod agent into a dev variant.
Export · Import
Portable JSON. Carry an agent across deployments, Git-track its config, restore from backup.
Can I ship without writing code?
Yes. Pick a marketplace template, walk the 7-step wizard, click publish. Or describe the agent in the AI Builder chat and let it draft the config. Ship in under an hour.
Can I express multi-step workflows visually?
The Workflow Designer ships 13 typed node types across structural, agent, tool, and control flow categories. Container nodes group sequential and parallel sub-workflows. Routers branch on conditions. The DAG validates before publish.
Do I have to use a proprietary SDK?
No. The Code Builder produces standard Python agents using open runtime primitives. The Katonic SDK is a thin resolver: tools.mcp() returns a binding, gateway.model() returns a binding. The runtime does the agent work. The SDK stays small and readable.
Do we have to migrate from LangChain?
No. BYOA accepts .py and .zip uploads, auto-detects the framework, extracts tools and model references, scaffolds the platform package around your code. You keep the framework. You gain governance, guardrails, observability, eval, and audit.
Only one of them lets every persona ship without a multi-quarter platform engineering project preceding the first agent.
Pick a builder, pick wrong.
Choose the SaaS console and your developers can't get under the hood. Choose the SDK and your business analysts can't ship without engineering. Choose either, lock half your team out.
You become the convergence.
Pick LangChain or CrewAI, build everything else yourself: guardrails, eval, observability, governance, multi-tenant audit. Months of platform engineering before the first agent ships.
Every persona, one runtime.
Five paths from no-code to BYOA. All converge on the same agent package. Same guardrails, governance, observability, and eval regardless of who built it. Marketplace head start. Source ownership.
Every team has its own preferred way to build. Forcing them all into one builder UX is how you end up with shadow agents on five different platforms. The right answer is to accept five entry points and produce one runtime, so the team gets to keep its habits and the platform gets to keep its governance.
Sandbox access in 24 hours. Office hours every Thursday with a Katonic engineer to walk a first agent end-to-end.
Join ePLDT, MODON, KAUST, Etisalat and enterprises across 11 countries.