Learning & Best Practices
This section collects hard-won lessons from production deployments of Open Astra. It covers common pitfalls, performance trade-offs, security hardening, multi-tenant architecture, and testing strategies — with real examples and actionable guidance.
Who this is for
These guides assume you've worked through the core docs and have at least one agent running. They're aimed at operators and developers taking Open Astra from prototype to production.
Guides
| Guide | What you'll learn |
|---|---|
| Anti-Patterns & Gotchas | Common mistakes — missing budget pre-flight, ignoring blackboard state, over-provisioning swarms. Real examples and fixes. |
| Performance Tuning | Trade-offs: batching tool calls vs. latency, memory tiers vs. retrieval accuracy, RRF weights vs. relevance. |
| Security Hardening | Step-by-step checklist: JWT configuration, secret rotation, workspace isolation, rate limiting, approval workflows. |
| Multi-Tenant Guide | Building a SaaS where customers have isolated agent teams — data, cost, and permission boundaries. |
| Testing Strategies | Unit testing agents, integration testing swarms, and e2e testing workflows. |
Core principles
- Constrain early. Set quotas, budgets, and approval gates before deploying to users — retrofitting guardrails is harder.
- Observe before optimizing. Instrument with Agent Metrics and read the data before tuning memory or inference settings.
- Test the boundaries. Agents behave differently under adversarial input, long sessions, and concurrent swarms — test all three.