The platform.
Every module gives you the capability, not the system you would otherwise build, secure and keep current. Start with one, add more as you grow.
AI
Agents
not an agent harness
Run agents with tools, memory and limits. Traces, retries and cost tracking come included.
Model routing
not a model gateway
Send each task to the right model, with fallbacks and budgets. EU-hosted models when your customers require them.
Knowledge
not knowledge base management
Bring in documents and data, keep them fresh, and choose presets for how agents use them.
Answers from your data
not a search stack
Retrieval that works out of the box. Need custom logic? Register your own retrieval and keep the rest.
Quality checks
not an evaluation pipeline
Evals that test agent output against real cases before and after every change, including a new model.
Approvals
not a human-in-the-loop system
Let people review, edit or reject what an agent or automation is about to do.
Automation
Automations
not an orchestration system
Multi-step work that survives failures, waits for events and picks up where it left off. Let your users build their own.
Scheduled work
not a job queue
Background and recurring tasks without running queues or cron servers.
Integrations
not a connector framework
Connect to the tools your customers already use, with auth and syncing handled.
Product foundations
Access rules
not an RBAC system
Roles, permissions and sharing that agents respect too. Define the rules once.
Workspaces
not a multi-tenancy layer
Organisations, teams and invites, with each customer’s data kept apart.
Notifications
not a notification system
Email, in-app and chat, with user preferences and standard frontend components.
An audit trail
not an audit log system
Who did what and when, whether it was a person or an agent. Ready for your customer’s security review.
Usage tracking
not a metering pipeline
Count what your customers use, per workspace, ready for limits and billing.
Operations
Alerts
not an alerting system
Know when something breaks, before your customers tell you.
Traces
not an observability stack
Follow a request through your app, your automations and your agents in one place.
How it is built.
Curated, not a menu
Less choice, better fit. You describe the product, we recommend the modules. When you need something new later, it gets added.
A hardened core
Extend every module through defined extension points. The core stays intact and tested, so nothing you or a coding agent writes can break it.
Serviced after launch
You own your product and its domain logic. We keep the modules under it stable, fast at scale, patched and upgraded.
Stable interfaces, swappable engines
The engine underneath a module can grow from a simple Postgres queue to Kafka without your code changing.
Frontier, opt-in
New models and patterns are evaluated once, centrally, and offered with a recommendation for your product. You choose when to move.
Compliance in the foundation
Audit trail, access rules and traceability are built in, so much of the evidence for NIS2, DORA and the AI Act comes with it. European infrastructure end to end, including EU-hosted models, when you need it.
Service without the agency
Much of the service is done by agents, not billable hours, and every test and fix benefits every team on Archipela.
Join the waitlist
We are opening Archipela to a small group of early teams first. Leave your email and we will be in touch.