Click any tag below to further narrow down your results
Links
Perplexity's new Context API lets AI agents pull up to 10 web pages at once and get back clean Markdown, HTML, links, and query-matched passages instead of raw HTML full of ads and navigation. Developers can control how fresh the content must be, from cached copies to live fetches.
- Returns clean Markdown with boilerplate like cookie banners, navigation, and ads stripped out, and can return only passages relevant to a query, so the model processes less text.
- Freshness is adjustable: by default it uses a cached copy from the last two days, but max_age_seconds set to 0 forces a live read, and cache_only prevents live requests entirely.
- In Perplexity's own evaluation, 92.3% of URLs returned usable content, 87.9% of key facts were recovered, and content precision was 80.3%.
Unblocked argues that giving an agent access to Slack, Jira, and GitHub through separate connectors isn't enough, because the agent still has to search each silo, reconcile conflicting answers, and guess at unwritten team conventions. Its benchmarks claim up to 83% faster task completion and about 40% fewer tokens with a context engine doing that work upfront.
- Per-source connectors return documents from one system, so cross-source questions like "why does the payment service retry like this?" force the agent to assemble answers itself at inference-time token cost.
- Connectors can't tell which source is current when a Confluence page, a later Slack thread, and a PR comment disagree; the article says a context engine resolves contradictions with recency and authority signals before answering.
- Unwritten conventions and expertise patterns exist only across systems over time, so no query can retrieve them. The article's benchmarks show a 9.5/10 convention-following score with context versus 2/10 without.
- Each direct connector adds its own credential and permission model, which the article presents as a security burden that a single server-side connection reduces.
The author tested whether agentic retrieval-augmented generation, where an LLM agent rewrites queries and decides what to fetch, actually finds better evidence than plain search and reranking. The whole setup is reproducible from a GitHub repo, with 298 reviewed questions and saved results.
- The test pits agentic RAG against plain search and a reranker on 298 questions, with Claude models writing the questions and running the agent.
- Results are reproducible: the code, questions, and saved outputs for every run are published, so you can check the numbers yourself.
- Reciprocal-rank fusion is the core merge step, combining several ranked lists of retrieved documents into one.
@0xMortyx on X
A data scientist ran a blind test pitting RAG against Kimi K3's full 1M-token context window on the same 12 questions. Long context scored perfectly on answer quality but cost about 16 times as much and ran roughly 3 times slower, so the post argues the right choice depends on corpus size and query volume.
- Long context answered all 12 questions perfectly, while RAG was near-perfect but weaker on completeness.
- Long context cost about 16 times as much and took roughly 3 times as long per question; at 12,000 queries the cost gap reaches about $3,800 versus $230.
- Skipping RAG is only backed by evidence when the corpus sits well under a fifth of the window; the test never covered corpora filling 60-80% of it.
- For tasks spanning many unconnected sources, the author points to Kimi K3's Agent Swarm and a connected context graph instead of either approach.
@0xMortyx on X
A 90-minute Google Cloud workshop walks through building production AI agents, moving from basic retrieval-augmented generation (RAG) to Graph RAG, memory, and multi-agent orchestration. The post shares a timestamped chapter guide and links to a written roadmap covering the same progression.
- The workshop is organized as a progression: RAG, then Graph RAG, then memory, then multimodal agent graphs.
- Hands-on sections cover setting up a production agent stack, Graph RAG with semantic search, extracting graph context, and orchestrating specialized agents.
- The author frames it as one of the clearer hands-on examples of teams moving beyond plain RAG toward graph-based context.
MemPalace is an open-source tool that stores and retrieves verbatim conversation history on your machine using semantic search and a pluggable vector-store backend. It supports ChromaDB, SQLite exact vectors, Qdrant, and pgvector, and ships a CLI and Docker images for easy setup without external APIs. Benchmarks show 96.6% recall at top-5 on LongMemEval with zero cloud or LLM calls.
- 96.6% recall at top-5 on LongMemEval with zero cloud or LLM calls, rising to 98.4% with hybrid keyword/temporal boosting and ≥99% with LLM rerank
- Runs fully local by default (verbatim storage, semantic search, ChromaDB/SQLite/Qdrant/pgvector options) so nothing leaves your disk unless you opt into a remote backend
- Also benchmarked at 60.3% R@10 on LoCoMo (88.9% with hybrid v5) and 80.3% R@5 on MemBench
- Ships as CLI, Docker (CPU/GPU) images, and an MCP server for direct integration with tools like Claude Code
Karpathy proposes replacing on-the-fly retrieval with an LLM-maintained markdown wiki that ingests sources, compiles structured pages, and self-updates through ingest, query, and lint cycles. This approach builds a persistent, compounding knowledge base without vectors or re-retrieval, though it currently lacks enterprise controls.
- Karpathy's "LLM Wiki" replaces vector-based RAG with an LLM that reads sources once and maintains a persistent, self-updating markdown wiki instead of re-retrieving chunked fragments per query.
- The system runs three operations—ingest (add new material), query (answer from the wiki, then save new answers back into it), and lint (scheduled scans for outdated info, broken links, and gaps)—so every interaction compounds the knowledge base rather than starting fresh.
- It directly targets two RAG failure modes: stateless retrieval that repeats embedding work every query, and chunking that destroys document context.
- The idea echoes Vannevar Bush's 1945 Memex concept of an associative, hyperlinked personal knowledge library, though it currently lacks enterprise-grade controls.