[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"$f1yi9km908pjgv":3},{"code":4,"message":5,"data":6,"timestamp":51},10000,"success",{"full_name":7,"position":8,"description":9,"language":10,"stars":11,"forks":12,"stars_gained":13,"age_days":14,"hn_threads":15,"one_liner":22,"summary_md":23,"topic":24,"github_topics":27,"license":47,"homepage":48,"updated_at":49,"history":50},"Graphicle-AI\u002Fgraphicle",22,"Code graph and memory for your coding agents, in one Rust program. For Claude Code, Codex, Cursor and any MCP client.","Rust",356,27,34,10,[16],{"id":17,"title":18,"points":19,"comments":20,"posted_at":21},50088713,"Show HN: Graphicle, a Rust port of Graphify with memory for coding agents",48,15,"2026-10-14","A local code graph plus project memory that serves cited excerpts to coding agents over MCP.","Worth a trial on a mid-sized repository where your agent spends turns grepping blindly; treat the speed figures as the author's claim and the answer-quality question as open.\n\n## The problem\n\nYou want a coding agent to find where failed requests are retried, or what a change to `retry()` touches. Today it guesses search terms, greps, reads whole files and often runs out of context before reaching the right function. Graphicle changes that step: setup indexes the repository and connects MCP clients, and the agent then requests excerpts ranked from code, docs, sessions and approved memories, each cited to a file and line range, so you can check the answer at the cited location.\n\n## What's different\n\nThe graph is built by tree-sitter parsing with no model, so indexing code needs no API key. Memory is the more interesting half: captured agent sessions and memories proposed by agents become sources, but proposed memories wait for your approval in the app. The README frames it as local-first: one SQLite store, loopback-only endpoints, no account or telemetry, with network calls tied to commands you name.\n\nThe trade-off is that the tools call no model themselves; answer quality still depends on the agent's model knowing how to consume structured excerpts, and the README says long sessions and memory built across sessions are not yet tested. The benchmark compares the memory engine's time against Graphify on 13 codebases, an author measurement rather than independent evidence of better answers. `graphicle disconnect` undoes what setup added and changed files are backed up first, but that is file restoration, not a promise every upgrade goes smoothly. The engine is explicitly a Rust port of Graphify, so if you already run Graphify the added value is mostly memory and single-binary packaging.\n\n## Who should try it\n\nGood fit: a developer on a substantial codebase with many cross-file calls, using Claude Code, Cursor, Codex or another MCP client, who wants project decisions and past session context to survive across sessions and will approve memories rather than trust them blindly. Also fine for someone who wants a local `ask` command and a browsable graph without a separate memory server.\n\nPoor fit: a small repo where plain ripgrep answers in one turn, or a team needing answer-quality evidence first. Code indexing is model-free, but a full `graphicle graph extract` sends documents and images to an LLM it finds a key for, so offline-only use is a different configuration from the default.\n\n## Requirements and limitations\n\nInstall is Homebrew, the install script or release archives. Windows is an x86-64 preview, and the signing policy says release builds are not yet signed with a trusted certificate, with Windows binaries on a test certificate or unsigned; check release notes there. Setup edits agent config and `AGENTS.md` and installs Git hooks, so you must accept file changes in the repo; backups and disconnect help, but keeping the repo current becomes upkeep rather than a one-time step. The daily update check contacts GitHub for the latest release by default; `GRAPHICLE_NO_UPDATE_NOTIFIER=1` disables it. The memory loop improves only if you review proposed memories, which is a habit rather than a feature.\n\n## HotRepo's take\n\nHotRepo first tracked the project on 2026-10-24; its only stored observation is #22 on the NEW-REPOSITORY board (rust) on 2026-10-25 with 356 total stars, a snapshot rather than growth. The Show HN thread had 48 points and 15 comments on 2026-10-14; one commenter reported that their non-rigorous evals had memory systems doing worse than a raw harness with ripgrep, and a reply argued such tools can help weaker models while slowing frontier ones. Persistence cannot be judged from two tracked dates. Verify by running its own agent benchmark on your repository against your normal grep workflow, per the [README](https:\u002F\u002Fgithub.com\u002FGraphicle-AI\u002Fgraphicle\u002Fblob\u002Fmain\u002FREADME.md).",{"slug":25,"name":26},"developer-tools","Developer Tools",[28,29,30,31,32,33,34,35,36,37,38,25,39,40,41,42,43,44,45,46],"agent-memory","ai-agents","claude-code","code-analysis","code-graph","code-intelligence","code-search","codex","coding-agents","context-engineering","cursor","graphrag","knowledge-graph","local-first","mcp","mcp-server","model-context-protocol","rust","tree-sitter","Apache-2.0","https:\u002F\u002Fwww.karlex.ai","2026-10-25T03:12:25Z",[],1792909370]