{"id":"2038709204769546692","url":"https://x.com/BniWael/status/2038709204769546692","text":"","author":{"name":"ProxySoul","username":"BniWael","avatarUrl":"https://pbs.twimg.com/profile_images/2047361525346648065/on50qptq_200x200.jpg"},"createdAt":"Mon Mar 30 20:05:26 +0000 2026","engagement":{"replies":3,"retweets":4,"likes":45,"views":4736},"article":{"title":"I Built an AI Coding Agent That Already Knows Your Codebase Before It Reads a Single File ","previewText":"Most AI coding agent starts the same way.\nYou give it a task, It greps, It reads a file, It greps again, reads three more files then Half its token budget gone, and it still doesn't understand the","coverImageUrl":"https://pbs.twimg.com/media/HErDQPTbAAA1UgO.jpg","content":"Most AI coding agent starts the same way.\n\nYou give it a task, It greps, It reads a file, It greps again, reads three more files then Half its token budget gone, and it still doesn't understand the full architecture.\n\nI spent months watching agents burn through context and time just to figure out where things are. Not because the models are bad. Because they're flying blind. They see files. They don't see systems.\n\n![](https://pbs.twimg.com/media/HErQ7SqXcAAxhpI.png)\n\nSo I built one that does.\n\n## What SoulForge Actually Does Differently\n\nWhen you open a project, SoulForge parses every file across 30+ languages, extracts every symbol and import edge, and builds a live dependency graph backed by SQLite.\n\nBefore the agent touches anything, it sees a Soul Map, every file ranked by importance, every exported symbol indexed, every dependency arrow drawn:\n\n![](https://pbs.twimg.com/media/HErRFZ5WEAAO0jw.png)\n\n- Which files matter most, PageRank ranks by structural importance, boosted by your recent edits\n\n- What changes together, git cochange analysis catches coupling that imports don't show\n\n- How far an edit ripples, blast radius tells the agent \"be careful, 30 files depend on this\"\n\n- Where everything is, full-text search over every symbol with signatures, answered in milliseconds\n\nThe graph updates in real-time. Edit a file, the map recalculates. The agent always sees the current state.\n\nThis isn't context stuffing. It's understanding.\n\n## Tools That Query the Graph, Not the Filesystem\n\nMost agents get grep and read_file and hope for the best. SoulForge gives the agent tools that answer from the graph directly, zero file reads, zero LLM tokens:\n\n- soul_find, \"where is AuthMiddleware?\" → PageRank-ranked results with signatures, instantly\n\n- soul_grep, checks identifier frequency in the graph before even spawning ripgrep\n\n- soul_analyze, unused exports, file profiles, top files, external packages\n\n- soul_impact, \"what breaks if I change this?\" → dependents, cochanges, blast radius\n\n![](https://pbs.twimg.com/media/HErRRElawAAZCr3.png)\n\nOne graph query replaces a dozen grep-read cycles. \n\n![](https://pbs.twimg.com/media/HErSCOvXkAA-e09.png)\n\nThe agent does less work, makes fewer mistakes, and you trust the result more.\n\n## Your Conventions, Loaded Once\n\nDrop a SOULFORGE.md in your project root. Architecture notes, naming conventions, toolchain preferences, and it's injected into every prompt.\n\nAlready using Claude Code, Cursor, or Copilot? SoulForge loads instruction files from 10 formats: CLAUDE.md, .cursorrules, AGENTS.md, .github/copilot-instructions.md, and more. Your conventions follow you across tools.\n\nPlus installable skills for domain-specific knowledge. The agent knows how your codebase works, not just what's in it.\n\n## 4-Tier Code Intelligence\n\nFor precise operations, SoulForge routes through four tiers: LSP → ts-morph → tree-sitter → regex.\n\nThe agent gets the full LSP surface, definitions, references, workspace rename, call hierarchy, diagnostics. When LSP isn't available, it falls through gracefully. Every language always gets the best available intelligence.\n\nThis powers compound tools that do the complete job in one call:\n\n- rename_symbol, LSP workspace rename with grep verification across all files\n\n- move_symbol, extract to another file, update all importers automatically\n\n- rename_file, rename/move with import path updates\n\n- refactor, extract function/variable, organize imports\n\nOne tool call. Not twelve.\n\n## Parallel Agents That Actually Coordinate\n\n![](https://pbs.twimg.com/media/HErRd-wXkAE8yOg.png)\n\nBig tasks fan out to parallel agents through a shared bus:\n\n- Investigate agents scan broadly, counts, patterns, structural analysis\n\n- Explore agents extract targeted context from specific files\n\n- Code agents edit with full findings already cached\n\nUp to 8 agents, 3 running concurrently. One agent reads a file, every other agent gets it instantly. Edit ownership prevents race conditions. A dependency DAG ensures code agents start with research already done.\n\nThe agent decides when to dispatch. Small tasks stay direct. Complex tasks fan out automatically.\n\n## Your Actual Neovim, Embedded\n\nThis was non-negotiable for me. I wanted to co-edit alongside the agent, not just watch a chat window.\n\nSoulForge embeds real Neovim via msgpack-RPC. Your config loads. Your plugins work. LazyVim, Mason, treesitter, Catppuccin, all of it. The AI and your editor share the same LSP connection. When the agent renames a symbol, you see it in the buffer. When you edit, the agent sees it.\n\nIt works over SSH. tmux, screen, remote servers. No Electron, no browser. Terminal-native.\n\n## 10 Providers, Per-Task Routing\n\nLLM Gateway, Anthropic, OpenAI, Google, xAI, Ollama (local), OpenRouter, Vercel AI Gateway, a local proxy, and any custom OpenAI-compatible API.\n\nPer-task model routing sends different tasks to different models automatically. Opus for coding accuracy. Haiku for cheap search queries. Sonnet for planning. Cost visible per task, per agent, per model.\n\nYou see exactly what you're spending.\n\n## Context That Doesn't Rot\n\n![](https://pbs.twimg.com/media/HErRw3HXsAAkdQZ.png)\n\nThis is where every AI tool breaks down. Hour-long sessions. Context fills up. Every API call sends 100K tokens of stale content.\n\nSoulForge tracks working state deterministically from tool calls. When context fills up: serialize, compress, continue. 80% token reduction. Zero LLM cost. Sessions save incrementally, crash, close, resume from exactly where you left off.\n\n## Safety That Isn't Optional\n\nApproval gates on every destructive action, rm -rf, git push --force, edits to .env or credentials. Individually prompted. No \"Always Allow\" button.\n\nForbidden files blocked across every tool. The LLM never sees your secrets. Pre-commit gate auto-runs lint and typecheck before any git commit.\n\n## The Numbers\n\n- 33  tools. \n\n- 30+ languages. \n\n- 10 providers. \n\n- Up to 8 parallel agents. \n\n- 23 auto-detected toolchains. \n\n- 21 themes. \n\n- 86 slash commands. \n\n- Headless mode for CI/CD.\n\nBSL 1.1 — free for personal and internal use.\n\n## Get Started\n\nHomebrew:\n\nBUN:\n\nThen just run: soulforge\n\nDocs: soulforge.proxysoul.com\n\nSource: github.com/ProxySoul/soulforge\n\nIf you've ever wished your AI coding agent understood what it was touching before it touched it, try it.\n\nDisclaimer: SoulForge is in early release, if you face any issues don't hesitate to raise an issue or contibute <3 \n\nProxySoul\n\n#ai #claudeCode #claude #CLI #Agent #Neovim #OpenCode"},"adhxContext":{"savedByCount":1,"publicTags":[],"previewUrl":"https://adhx.com/BniWael/status/2038709204769546692"}}