Ujex Blog
Posts on every layer of agent infrastructure.
- Open Source AI Agent SDK Comparison 2026: LangGraph, Letta, Mastra, Agno, Pydantic AI, UjexSix open-source agent SDKs compared by what they actually do. Frameworks vs runtimes vs substrates. With code samples and a clear pick-by-use-case guide.Apr 12, 2026
- How to Give an AI Agent an Email Address (2026 Guide)How to give your AI agent its own real email inbox — with or without a SaaS vendor. Six approaches compared (AgentMail, Nylas, Mailgun, self-hosted Postfix, Composio, Ujex Postbox) with code samples and cost.Apr 05, 2026
- Self-Hosting Postfix + OpenDKIM + Brevo for AI Agent EmailThe runbook. Hetzner VM, Postfix on :25 (when port-25 unblocks), OpenDKIM signing, Brevo as the relay until you don't need it. End-to-end DKIM/SPF/DMARC. ~4 hours, $5/month.Mar 19, 2026
- Building With LangGraph On Top Of Ujex (Reference Architecture)LangGraph for the orchestration. Ujex for the substrate (email, memory, ingress, audit, mobile). Concrete reference architecture with code, deploy commands, and where each piece runs.Feb 24, 2026
- Best Email APIs for AI Agents in 2026 (Honest Comparison)Six email APIs, ranked by use case. AgentMail, Nylas, Composio, Mailgun, self-host, Ujex Postbox. With actual costs at 1k / 10k / 100k messages/month.Feb 12, 2026
- Git-Backed Memory for AI Agents: Why Markdown Beats Vector StoresYour agent's memory should live in git, not a database. Here's the case for markdown-first, bucket-as-source-of-truth memory — version control, grep, diff, no vendor lock-in.Jan 30, 2026
- Mem0 vs Letta vs Zep vs Ujex Recall: Picking an Agent Memory LayerFour agent memory products. One isn't an agent framework. One is a managed graph memory. One is markdown-first and lives in your bucket. Honest matrix + when each is right.Jan 25, 2026
- Human-in-the-Loop AI Agents: A Production Setup GuideAdding approval gates to autonomous AI agents. Mobile-first design, sub-second latency, biometric auth. Code, architecture, and the design decisions that matter at scale.Nov 26, 2025
- LangGraph + Ujex Recall: Long-Term Memory Without Building Your OwnDrop Ujex Recall into a LangGraph agent loop. Markdown-first memory in your bucket; BM25 + vectors. Survives across sessions, persists across deployments, greppable on your laptop.Oct 22, 2025
- Self-Hosted vs SaaS Agent Infrastructure: When Each Makes SenseSelf-hosted AI agent stack vs cloud provider comparison. Data residency, compliance risks, vendor lock-in, ops cost. When self-host wins; when cloud wins; when hybrid is right.Sep 30, 2025
- A Hash-Chained Audit Log for Your AI Agent (And Why It Matters)When your agent does something weird, can you say what it did? Append-only, hash-chained logs are the difference between investigation and guesswork. Code, structure, and a 200-line library.Sep 03, 2025
- Migrating From Mem0 / Letta / Zep to Ujex RecallExport, transform, re-ingest. Plus the things to think about: embedding model differences, namespace mapping, and what to do with frontmatter.Aug 31, 2025
- AgentMail Alternatives: 5 I Actually TestedI built test agents on five AgentMail alternatives: Nylas, Mailgun + custom, Composio, self-host Postfix + Brevo, Ujex Postbox. Here's what each is good at and where each falls down.Aug 22, 2025
- Stopping a $5,000 Surprise Bill From An AI AgentHow to stop your AI agent from running up a huge API bill automatically. Per-agent spending limits with real-time alerts. Open-source governor for AI agent cost management.Jul 14, 2025
- Cloudflare Tunnel vs ngrok vs Ujex IngressThree reverse-tunneling stacks. Different sweet spots. Cloudflare Tunnel for free + custom domain. ngrok for fast prototyping. Ujex Ingress for agent-shaped workloads.Jun 19, 2025
- Eight Vendors Or One: The Real Cost of Glue InfrastructureAn opinion piece. Most agent teams are running SendGrid + ngrok + Pinecone + Temporal + Auth0 + Twilio + Stripe + Honeycomb. Eight vendors, eight integrations, eight failure modes.May 22, 2025
- MCP Server for Email: Adding Postbox to Claude DesktopWire Ujex Postbox into Claude Desktop / Cursor / Cline / Zed via the @ujex/postbox-mcp server. Five minutes; agent gets full inbox/outbox tools.May 19, 2025
- AI Agent Audit Logs for HIPAA, SOC 2, and GDPR ComplianceWhat auditors actually need from your agent audit trail. Hash chains, retention, PII handling, and the gap between 'we have logs' and 'we have an audit log that passes review.'May 10, 2025
- Phone Approvals For An AI Agent: How `mobile.ask` WorksHuman-in-the-loop AI agent setup using mobile push. Approve before send workflow for autonomous agents — sub-second mobile oversight, biometric confirm, FCM-backed. The mobile.ask primitive in detail.Apr 30, 2025
- How to Receive a Webhook on a Local AI Agent (7 Ways, Honest)Your agent runs on your laptop. OpenAI / Stripe / GitHub / your own service needs to call it. Here are seven ways, with code samples and tradeoffs for each.Apr 22, 2025
- What Even Is 'Agent Infrastructure'? A Field Guide for 2026Agent frameworks, agent runtimes, agent infra, agent observability, agent evals, agent memory layers — the words overlap. This is a definition pass with examples in each box.Apr 19, 2025
- Why We Built Postbox When AgentMail ExistsHonest founder post. AgentMail is a great product. Here's why we built Postbox anyway — and why it makes sense to have both options in the world.Mar 15, 2025