In your system
Magnus API
Your own app, website or CRM talks to the agent through an OpenAI-compatible API and SDKs. The interface and the brand stay yours.
You are here
Magnus API · in your own system
AI that actually executes. Without losing control.
The same Magnus agents, inside your own app: an OpenAI-compatible API and SDKs, with rules that decide, confirmation before anything irreversible and a trace of every turn.
You stay in control — define what can and can’t happen
Human in the loop — approve critical actions
Fully auditable — every decision, every step
Design
Execute
Observe
Scale
The Shift to Execution
Magnus replaces fragmented AI calls with a unified agent execution runtime.
Legacy Web Apps
- Isolated prompt-response calls
- Fragile custom glue logic
- Deterministic paths require hardcoding
- No state awareness between interactions
- Black-box decision making
Agent Execution Runtime
- Active decision-driven agents
- Engine decides the optimal next step
- Native execution of system actions
- Persistent state and execution context
- Fully auditable reasoning traces
"Magnus doesn't respond to prompts. Magnus executes processes."
Execution Transformation
Unlock decision-driven
agents in 3 steps
Move beyond simple responses to agents that execute real actions and manage state — without changing your stack.
01
1. Run
Point your OpenAI client at Magnus: a new base URL and your Magnus key.
02
2. Decide
Set the behavioral constraints and decision-making logic for your agent.
03
3. Act
Run agents that execute real actions across your systems.
Standard Outputbase_url: "https://app.iamagnus.com/v1"
Works with the official OpenAI SDKs. Tools live in the agent, not in the request.Real Execution in Action
See how a Magnus agent executes a complex IT Support workflow within defined constraints.
Execution Trace: IT-HELPDESK-082
LIVE RUNTIME
Input Received
"Hi, I'm locked out of my account and I have an urgent meeting in 10 minutes. Can someone help?"
Decision Engine Assessment
Policy CheckKnowledge RetrievalAction Selection
Engine identified high-priority account lockout. Retrieving security constraints and selecting optimal recovery action.
Action Executed
POST /api/v1/auth/reset-lockout
SUCCESSContinuous Observation
Confirming lockout status was cleared. Verifying if a new password was sent via secondary email.
Workflow completed · trace recorded
What Magnus API is
Infrastructure to operate AI, not another LLM wrapper.
What it IS
- A decision engine at its core
- A centralized LLM Gateway
- A policy and security engine
- A memory and context layer (RAG)
What it is NOT
- A generic 'no-code' chatbot
- A simple OpenAI wrapper
- An isolated vector database
- A UI design tool
Runtime Capabilities
Native agent execution. Everything you need to scale agent-driven operations.
True Multi-tenancy
Data, keys and agents isolated per organization.
Execution Gateway
Automatic failover between OpenAI and Google Gemini. On paid plans, bring your own provider key.
Decision Engine
Deterministic engine that operates within defined rules and coordinates cross-system actions.
Knowledge & Context
Shared knowledge injection from your private systems. Exact citations and state-aware retrieval.
Persistent State
Conversation state kept on the server: each end user's conversation continues where it left off.
System Actions
Ready connectors for Jira and SAP (read-only), plus your own APIs and MCP servers.
Outcome Engine
Transforms agent reasoning into structured execution over real systems.
Human Loop
Smart handoff to human operators when the agent detects ambiguity — maintaining full state context.
Full Observability
Per-turn traces and token usage. Every agent decision is auditable.
Built for real-world execution
Security and trust at every layer of the product.
No black boxes
No hidden decisions, everything is transparent.
Full traceability
All actions are auditable in real-time.
Role-based permissions
Granular control over who can do what.
Controlled autonomy
AI operating on your terms, always.
One agent, two ways to use it
Don't want to write code?
It is the same agent, with the same rules and the same plan. All you choose is where it serves.
In your channels
Magnus Agents
The same agent serves on WhatsApp, Telegram and your website chat, set up from the dashboard. No code.
See Magnus Agents →Paid plans include both. See plans
Resources
Frequently Asked Questions
LLMs are probabilistic and change over time. Magnus transforms them into deterministic tools. We wrap the model's intelligence with a 'Decision Engine' that ensures business rules are always met, regardless of the LLM's variability.
LangChain is great for low-level prototyping. Magnus is an agent execution runtime designed for enterprise production. We prioritize predictability and deterministic decisions over experimental autonomy.
Agents can interact with Jira, schedule appointments, manage tickets, and call any REST API. They also connect to external MCP (Model Context Protocol) servers.
Still have questions?
Write to us and the team will answer.
Stop experimenting with AI. Start executing with control.
Build reliable agents today.
OpenAI-compatible API
Included in every paid plan