AI Agent Memory Types: In-Context vs Episodic vs RAG vs Semantic
In-context, episodic, RAG, and semantic memory for AI agents on Spring Boot. Real ContextService from production, decision tree, and code with pgvector.
Useful articles about Java, Spring, SEO, frontend, and modern technologies. Tips, examples, and lifehacks for developers
In-context, episodic, RAG, and semantic memory for AI agents on Spring Boot. Real ContextService from production, decision tree, and code with pgvector.
Grok Build by xAI: Plan Mode, 2M context tokens, parallel sub-agents. Technical review of the early beta CLI agent. Comparison with Claude Code and Codex CLI.
Ollama adds official support for OpenAI Codex App. Run a powerful local AI coding agent on any Ollama model with one command — no OpenAI subscription required.
After 10-15 tools, selection accuracy drops. RAG tool solves this through vector search of the tool registry. Implementation on Spring AI + pgvector with code and numbers
Empty tool result, low relevance score, API error — how your agent hallucinates without grounding and how to fix it. Confidence scoring + re-query in Spring AI.
Я очікував що AI здасться через 3 раунди. Він не здався через 8. І це змінило моє розуміння того як працюють мовні моделі. Як виникла ідея Класична проблема AI-агентів — вони занадто ввічливі. Попроси ChatGPT посперечатись — він погодиться через два повідомлення. Мене це дратувало. Я...
How to build a multi-agent system on Spring AI: @Async dialogue loop, switching Ollama and OpenRouter via @Profile, five tools and prompts that make agents
GPT-Realtime-2 vs Gemini Live API compared: pricing, benchmarks, video, SIP, languages. 6x cost gap — and which one fits your use case. Updated May 2026.
GPT-5.5 in Codex: 82.7% on Terminal-Bench, ~40% fewer tokens per task, new Fast mode. Comparison with GPT-5.4, limitations, and practical developer experience.
Step-by-step guide to GPT-Realtime-2 Realtime API: WebSocket vs WebRTC vs SIP, working code in JS and Python, preambles, tool calls, common pitfalls. Updated May 2026.
OpenAI released GPT-Realtime-2, Translate, and Whisper. What has changed, real Zillow and Deutsche Telekom figures, prices, and why OpenRouter won't work.
Which Ollama models actually support tool calling in 2026: comparison of qwen3, llama3.1, gemma4, mistral-nemo. Benchmarks, reliability table, common errors
GPT-5.3-Codex-Spark — the first real-time Codex model: >1000 tokens/sec on Cerebras. How it differs from GPT-5.5, how to enable in Codex App
OpenAI Codex у 2026 році — це не той інструмент, про який ви, можливо, читали кілька років тому. Оригінальний Codex API (2021–2023) був моделлю для автодоповнення коду на базі GPT-3, яка живила ранні версії GitHub Copilot. OpenAI закрила той API у березні 2023 року. Те, що існує сьогодні —...
Full guide to Ollama API: /api/chat, streaming, embeddings, tool calling. Examples in Java (WebClient + Spring Boot), Python, and JavaScript with working code.
Honest breakdown: where Ollama wins on privacy and cost, where ChatGPT and Claude pull ahead. Decision matrix, 2026 pricing, and a hybrid workflow that works.
DeepSeek V4 Pro — 1.6T parameters, MIT license, $3.48/M output vs $25/M for Claude Opus 4.7. We analyze the architecture, real benchmarks, where Pro wins, where it loses
deepseek-chat and deepseek-reasoner will be discontinued on July 24, 2026. Risk matrix, migration timeline, and a 15-minute checklist for technical managers.
$285B wiped in 48 hours. Prompt engineering is dead. Solo founders hit $1M ARR. A practitioner's analysis of what GPT-5.5 really means for SaaS, devs, and startups.
Benchmarks, real migration costs, and a decision checklist: where GPT-5.5 wins, where GPT-5.4 is still enough, and how to A/B test before you commit.
DeepSeek V4 Flash — 284B MoE, 1M context, $0.14/M tokens. Full review of architecture, benchmarks, and deployment via Ollama Cloud and DeepSeek API. From a practitioner.
Tested Claude Opus 4.7 on 400 legal PDFs in my RAG system AskYourDocs. Compared with Llama 3.3 70B — what wins, what costs, when to choose.
Claude Opus 4.7 released April 16, 2026. SWE-bench Pro 64.3%, vision 3.75 MP, price $5/$25. Analyzing what's new, breaking changes, and real-world experience in RAG.
The truth about Gemma 4 26B MoE that isn't written in promotional reviews. Swapping on Mac 24 GB, 2 tokens/sec, Flash Attention bugs.
What is thinking mode in Gemma 4, how it works through Ollama, how to shorten or disable it via /no_think. Real tests on M1 16 GB