MAGIC AGENT PLATFORM
The agent platform Magic's own operations run on.
Orchestration, instruction governance, context assembly, and memory — live with two paying customers, in production internally for two years. Available to design partners.
Start a conversation →The agent infrastructure in the market is either a toolkit you assemble yourself — dozens of components, no governance, your data inside a hyperscaler — or a vertical product handling one use case. Neither works for a mid-market organisation that needs production-grade agents on infrastructure they control. We built this for our own operations, proved it on two live customers, and now offer it to companies who want the same discipline applied to theirs.
THREE DISCIPLINES
01
Orchestration.
Which agent runs, when, on what data, with what authority. Every automated action is condition-checked at execution time. If the premise is no longer true, the action self-cancels with a documented reason.
02
Instruction governance.
What the agent is allowed to say, in whose voice, under whose authority. Structural prompts are versioned separately from dynamic knowledge. Magic's reasoning evolves independently of each customer's data.
03
Context assembly.
What the agent knows at the moment of reasoning. Per-request context from entity data, performance history, interaction timeline, and aggregate patterns. Assembled at constant token cost through materialised views.
IDENTITY
Identity is the root, not the wrapper.
Passwordless, cryptographic, everywhere.
Every person, organisation, and agent authenticates through cryptographic identity. No passwords. No credentials to rotate. One identity, verified once, trusted across the operations chat, the campaign engine, the community AI, and the attribution layer.
Same agent, different behaviour, enforced in code.
A community agent serves three tiers — admin, member, public — from the same interface. An organisation's chat gates actions by trust level. The boundaries are architectural. They do not depend on the model following instructions correctly.
MEMORY
An agent without memory is a chatbot. An agent with governed memory is infrastructure.
Semantic memory.
Natural-language addressable. Ingests unstructured content — emails, documents, transcripts — through embeddings. Queryable by similarity. Answers 'what did this customer say about pricing last time?' without loading every prior conversation.
Structured memory.
Schema-addressable. Queryable through typed schemas. Pipeline rows, contact records, campaign definitions, order history. Answers 'all open offers over €5K, sorted by follow-up date' with precision.
Memory is governed at the entity level. Your leadership controls what enters, what leaves, and what is exposed to which identity. Memory is owned by the entity. Magic operates the agent that reasons over it.
CONFIGURABILITY
Everything that varies between customers clusters into seven dimensions.
D1
Domain vocabulary.
What the organisation calls its entities and stages. Stored in config, read by all systems.
D2
Email transport.
Gmail, Microsoft 365, or SMTP. A unified send function routes by transport type.
D3
Selling motion.
Step count, cadence, channels, personalisation depth. Lives in the data, not the code.
D4
Product catalogue.
Tab name, column layout, grouping logic. A loader function abstracts the specifics.
D5
Team and trust levels.
Who can approve, log, view, create. Stored in configuration, enforced in code.
D6
Brand.
Colours, fonts, logo, tagline. CSS variables in frontends, string interpolation in emails.
D7
Identity and hosting.
Subdomain, agent identity, runtime, governed data store. Each customer gets their own isolated instance.
DATA RESIDENCY
We name what the architecture guarantees and what it does not.
What we guarantee today
Your data lives in your own cloud workspace — not in ours.
Magic does not hold or centralise customer data.
Runtime data path runs between your workspace and the AI layer — never through Magic's servers.
Revoke our access: immediate, complete loss of access.
Every schema change versioned and auditable.
Every automated action condition-checked at execution time.
Member-only content enforced in code, not in prompts.
What we do not currently guarantee
Compute runs on a cloud provider's multi-tenant runtime, not yet in a trusted execution environment you control.
The cloud provider encrypts data at rest; customer-managed encryption keys are not yet available.
Magic has operational access during the engagement — revocable at any time.
These gaps define our Horizon Europe research agenda, where formal verification of the Private AI pattern is a funded objective.
DESIGN PARTNERSHIPS
What a design partnership offers.
Direct working relationships with the team that builds the platform. Milestone-based engagements with preferential pricing. For organisations that want to run their own operations on Magic Agent Platform infrastructure. Not self-serve.
If this is the shape of what you want to build on, tell us what you have in mind.
Start a conversation →info@magic-id.com
We respect your privacy
We only use one cookie to remember your language preference. We don't use tracking cookies. No analytics, no tracking, no data collection — just a better experience for you.