10 articles found
A curated map of the OpenCode ecosystem in 2026: 10 persistent-memory plugins no blog has shown you, 4 real Hacker News critiques, and the indie community's defensive response.
Four visual clients for OpenCode put through real work on desktop and mobile. We compare install, day-to-day feel, remote access setup, and the parts each one still falls short on, with primary data from their own repos.
Route each subagent to the right model: explore to 9B, planner to 122B. Brick, OpenCode-subagent-router, and a DIY hook compared with real cost math.
RTK and Caveman promise 60-90% token cuts in agents. I measure both with real data: JetBrains' 8.5% finding, a 614M-token replay and the output-vs-prose gap.
HyperFrames vs Remotion 2026: the definitive technical comparison for AI agent video. React vs HTML, licensing, benchmarks, and Agent Skills.
Harness Engineering explained: why the wrapper (tools, memory, guardrails) makes agents productive. Definition, anatomy, and real indie application.
Comparative technical analysis of three native OpenCode plugins to give your AI agent persistent local memory: simple-memory (logfmt), Mnemosyne (offline Go binary), and true-mem (cognitive psychology).
Exhaustive technical comparison of three cross-platform MCP servers to give AI agents persistent memory: opencode-supermemory (cloud), basic-memory (Markdown + graph), and forgetful (atomic Zettelkasten). Works with Claude Code, Codex, Cursor and more.
Deep dive into Headroom, the open-source project (33.4k ⭐) that promises 60-95% token reduction on what you send to an LLM without losing accuracy. We test its 6 algorithms, its reversible CCR, and its proxy mode, and compare it to the rest of the ecosystem.
Deep dive into GSD Core (formerly Get Shit Done): 33 agents, wave execution, 200K fresh context per subagent, and the Discuss→Plan→Execute→Verify→Ship phase loop. How to install it, when to use it, and when not to.
