{"id":8305,"date":"2026-04-26T09:40:00","date_gmt":"2026-04-26T01:40:00","guid":{"rendered":"https:\/\/meta-quantum.today\/?p=8305"},"modified":"2026-04-26T10:56:16","modified_gmt":"2026-04-26T02:56:16","slug":"superintelligence-why-the-future-of-ai-is-a-file-system-coral","status":"publish","type":"post","link":"https:\/\/meta-quantum.today\/?p=8305","title":{"rendered":"SuperIntelligence: Why the Future of AI is a File System (CORAL), Setup &amp; Implementation Guide inside."},"content":{"rendered":"\n<h2 class=\"wp-block-heading\">\ud83d\udccc Introduction<\/h2>\n\n\n\n<p>A widely held assumption: that making AI smarter means training smarter models. The presenter argues that the AI research community has quietly pivoted in a different direction entirely \u2014 not improving the LLM itself, but engineering an increasingly sophisticated <em>environment<\/em> around it. The centerpiece of this argument is <strong>CORAL<\/strong> (Collaborative Open-ended Reinforcement Agent Learning), a new autonomous multi-agent infrastructure published jointly by MIT, Stanford, NUS, McGill, Meta, Microsoft, and Amazon. Rather than calling this progress &#8220;AI,&#8221; It coins the term <strong>ADI \u2014 Advanced Intelligence<\/strong> \u2014 to describe intelligence that lives in the harness, not the model. <a href=\"#video\" title=\"\">Watch the video<\/a>. Here is the <a href=\"#guide\" title=\"\">Complete Setup &amp; Implementation Guide.<\/a><\/p>\n\n\n\n<h2 class=\"wp-block-heading\">\ud83c\udd95 New Features and Core Concepts<\/h2>\n\n\n\n<p><strong>CORAL<\/strong> is described not as a mere agent framework but as a <em>complete technological infrastructure<\/em> for self-evolving multi-agent systems. Several key innovations stand out:<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">1. Hierarchical File System as Shared Memory<\/h3>\n\n\n\n<p>CORAL replaces the concept of internal model memory with a <strong>standardized, persistent, shared file system<\/strong>. All agents \u2014 typically 4 to 8 running in parallel \u2014 write to and read from a common <code>.coral\/public<\/code> directory. Three artifact types form the backbone:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>Attempts<\/strong> \u2014 JSON logs of every function evaluation, keyed by git commit hash<\/li>\n\n\n\n<li><strong>Notes<\/strong> \u2014 Markdown files where agents write textual hypotheses about the problem space<\/li>\n\n\n\n<li><strong>Skills<\/strong> \u2014 Reusable, abstracted code modules extracted from successful runs<\/li>\n<\/ul>\n\n\n\n<h3 class=\"wp-block-heading\">2. Isolated Git Worktrees Per Agent<\/h3>\n\n\n\n<p>Each agent operates on its own <strong>local git worktree<\/strong>, ensuring its exploratory changes don&#8217;t corrupt the shared global state. Coordination is achieved entirely through the shared file system \u2014 no message queues, no centralized orchestrator.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">3. Heartbeat Intervention Protocol<\/h3>\n\n\n\n<p>To prevent agents from getting trapped in local minima (a key failure mode in optimization), CORAL introduces a <strong>background asynchronous heartbeat<\/strong>:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>Interval trigger<\/strong> \u2014 After N evaluations, forces each agent to synthesize its notes into abstracted skills, effectively rewriting the starting point.<\/li>\n\n\n\n<li><strong>Plateau trigger<\/strong> \u2014 Detects stagnation and commands agents to attempt a <em>mathematically orthogonal approach<\/em>, conceptually equivalent to <strong>thermal noise injection in simulated annealing<\/strong>.<\/li>\n<\/ul>\n\n\n\n<h3 class=\"wp-block-heading\">4. Gradient-Free, Inference-Time Operation<\/h3>\n\n\n\n<p>Critically, no model weights are ever updated. Claude Opus 4 and Minimax M2.5 are used as frozen reasoning engines. All &#8220;learning&#8221; is <strong>in-context memory accumulation<\/strong> via the file system \u2014 making CORAL a <strong>gradient-free search algorithm operating entirely at inference time<\/strong>.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">5. Custom Task Grader<\/h3>\n\n\n\n<p>CORAL requires a task-specific <code>evalgrader.py<\/code> that functions as a <strong>reward function<\/strong> \u2014 returning a score (Boolean or scalar) for each agent submission, replacing the need for a human-in-the-loop or LLM-as-judge.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">6. CORAL.md Agent Instruction File<\/h3>\n\n\n\n<p>Each agent&#8217;s workspace contains a <code>CORAL.md<\/code> file \u2014 the deterministic instruction prompt defining the agent&#8217;s orientation workflow, ground rules, and behavioral constraints. The presenter draws a direct parallel to <code>CLAUDE.md<\/code> files in Claude Code&#8217;s ecosystem.<\/p>\n\n\n\n<p>Here is a comprehensive, step-by-step guide to building the CORAL infrastructure from scratch, sourced directly from the official repository and paper.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"guide\">\ud83e\udeb8 CORAL: Complete Setup &amp; Implementation Guide<\/h2>\n\n\n\n<p><strong>Repository:<\/strong> <a href=\"https:\/\/github.com\/Human-Agent-Society\/CORAL\">https:\/\/github.com\/Human-Agent-Society\/CORAL<\/a> <br><strong>Paper:<\/strong> <a href=\"https:\/\/arxiv.org\/abs\/2604.01658\">https:\/\/arxiv.org\/abs\/2604.01658<\/a><\/p>\n\n\n\n<h3 class=\"wp-block-heading\">Part 1 \u2014 Understanding the Architecture First<\/h3>\n\n\n\n<p>Before touching a single file, you need to internalize the mental model. CORAL is an infrastructure for building organizations of autonomous AI agents that run experiments, share knowledge, and continuously improve solutions. Give it a codebase and a grading script, and CORAL handles the rest: isolated workspaces, safe evaluation, persistent shared knowledge, and multi-agent collaboration to enable robust evolution.<\/p>\n\n\n\n<p>The entire system rests on <strong>five interlocking concepts:<\/strong><\/p>\n\n\n\n<pre class=\"wp-block-code\"><code>\u250c\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510\n\u2502                  CORAL ARCHITECTURE                 \u2502\n\u2502                                                     \u2502\n\u2502  \u250c\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510  \u250c\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510  \u250c\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510           \u2502\n\u2502  \u2502 Agent 1  \u2502  \u2502 Agent 2  \u2502  \u2502 Agent N  \u2502  (frozen  \u2502\n\u2502  \u2502 (Claude  \u2502  \u2502 (Claude  \u2502  \u2502 (Claude  \u2502   LLMs)   \u2502\n\u2502  \u2502  Code)   \u2502  \u2502  Code)   \u2502  \u2502  Code)   \u2502           \u2502\n\u2502  \u2514\u2500\u2500\u2500\u2500\u252c\u2500\u2500\u2500\u2500\u2500\u2518  \u2514\u2500\u2500\u2500\u2500\u252c\u2500\u2500\u2500\u2500\u2500\u2518  \u2514\u2500\u2500\u2500\u2500\u252c\u2500\u2500\u2500\u2500\u2500\u2518           \u2502\n\u2502       \u2502             \u2502             \u2502                 \u2502\n\u2502       \u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2534\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518                 \u2502\n\u2502                     \u2502                               \u2502\n\u2502             \u250c\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u25bc\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510                   \u2502\n\u2502             \u2502  .coral\/public\/   \u2502  (shared memory)  \u2502\n\u2502             \u2502  \u251c\u2500\u2500 attempts\/    \u2502                   \u2502\n\u2502             \u2502  \u251c\u2500\u2500 notes\/       \u2502                   \u2502\n\u2502             \u2502  \u2514\u2500\u2500 skills\/      \u2502                   \u2502\n\u2502             \u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518                   \u2502\n\u2502                     \u2502                               \u2502\n\u2502             \u250c\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u25bc\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510                   \u2502\n\u2502             \u2502  Grader \/ Eval    \u2502  (reward signal)  \u2502\n\u2502             \u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518                   \u2502\n\u2502                     \u2502                               \u2502\n\u2502             \u250c\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u25bc\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510                   \u2502\n\u2502             \u2502  Heartbeat Mgr    \u2502  (anti-stagnation)\u2502\n\u2502             \u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518                   \u2502\n\u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518\n<\/code><\/pre>\n\n\n\n<p>Each agent runs in its own git worktree branch. Shared state (attempts, notes, skills) lives in <code>.coral\/public\/<\/code> and is symlinked into every worktree \u2014 agents see each other&#8217;s work in real time with zero sync overhead. The manager watches for new attempts and can interrupt agents with heartbeat-triggered prompts (e.g. &#8220;reflect&#8221;, &#8220;consolidate skills&#8221;).<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">Part 2 \u2014 Prerequisites &amp; System Requirements<\/h3>\n\n\n\n<h4 class=\"wp-block-heading\">2.1 System Requirements<\/h4>\n\n\n\n<figure class=\"wp-block-table\"><table class=\"has-fixed-layout\"><thead><tr><th>Requirement<\/th><th>Detail<\/th><\/tr><\/thead><tbody><tr><td>OS<\/td><td>Linux \/ macOS (tmux required)<\/td><\/tr><tr><td>Python<\/td><td>3.11 or higher<\/td><\/tr><tr><td>Package manager<\/td><td><code>uv<\/code> (from Astral)<\/td><\/tr><tr><td>Git<\/td><td>Required for worktree isolation<\/td><\/tr><tr><td>API Key<\/td><td>Anthropic (Claude Code), OpenAI (Codex), or local (OpenCode)<\/td><\/tr><tr><td>Budget awareness<\/td><td>Single 4-agent run can cost up to <strong>$240 USD<\/strong> \u2014 set API spending limits before starting<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n<h4 class=\"wp-block-heading\">2.2 Install <code>uv<\/code> (Package Manager)<\/h4>\n\n\n\n<pre class=\"wp-block-code\"><code>curl -LsSf &lt;https:\/\/astral.sh\/uv\/install.sh&gt; | sh\n<\/code><\/pre>\n\n\n\n<p>Verify:<\/p>\n\n\n\n<pre class=\"wp-block-code\"><code>uv --version\n<\/code><\/pre>\n\n\n\n<h4 class=\"wp-block-heading\">2.3 Install and Authenticate Your Coding Agent<\/h4>\n\n\n\n<p>Before using CORAL, make sure you have fully set up the agent(s) you plan to use. Install the Agent following official installation instructions (e.g., Claude Code, Codex, OpenCode). Authentication: Login and authenticate your coding agent first to make sure they do not ask for your credentials in CLI mode. Set up any required environment variables, configuration files, or authentication secrets. CORAL does not handle agent installation or authentication for you. The infrastructure will fail to function if the underlying agent cannot start or is not properly authenticated.<\/p>\n\n\n\n<p><strong>For Claude Code (recommended, most tested):<\/strong><\/p>\n\n\n\n<pre class=\"wp-block-code\"><code>npm install -g @anthropic-ai\/claude-code\nclaude login\n<\/code><\/pre>\n\n\n\n<p>Then configure permissions so the agent can run autonomously. Edit <code>~\/.claude\/settings.json<\/code>:<\/p>\n\n\n\n<pre class=\"wp-block-code\"><code>{\n  \"permissions\": {\n    \"allow\": &#91;\"Bash\", \"Read\", \"Write\", \"Edit\"],\n    \"deny\": &#91;\"WebFetch\", \"WebSearch\"]\n  }\n}\n<\/code><\/pre>\n\n\n\n<h3 class=\"wp-block-heading\">Part 3 \u2014 Install CORAL<\/h3>\n\n\n\n<pre class=\"wp-block-code\"><code>git clone &lt;https:\/\/github.com\/Human-Agent-Society\/CORAL.git&gt;\ncd CORAL\n\n# Sync dependencies (Python packages via uv)\nuv sync\n\n# Optional: include web dashboard dependencies\nuv sync --extra ui\n<\/code><\/pre>\n\n\n\n<p>Verify the install:<\/p>\n\n\n\n<pre class=\"wp-block-code\"><code>uv run coral --help\n<\/code><\/pre>\n\n\n\n<p>You should see all 17+ CLI commands listed.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">Part 4 \u2014 Building Your First CORAL Task (Full Walkthrough)<\/h3>\n\n\n\n<p>The official TSP (Travelling Salesman Problem) example is the cleanest path to understanding the complete lifecycle.<\/p>\n\n\n\n<h4 class=\"wp-block-heading\">Step 1 \u2014 Create the Task Directory Structure<\/h4>\n\n\n\n<pre class=\"wp-block-code\"><code>mkdir -p examples\/tsp\/seed\nmkdir -p examples\/tsp\/eval\n<\/code><\/pre>\n\n\n\n<p>Your target layout:<\/p>\n\n\n\n<pre class=\"wp-block-code\"><code>examples\/tsp\/\n\u251c\u2500\u2500 seed\/\n\u2502   \u2514\u2500\u2500 solution.py      \u2190 agents iterate on THIS\n\u251c\u2500\u2500 eval\/\n\u2502   \u2514\u2500\u2500 grader.py        \u2190 scores agent submissions\n\u2514\u2500\u2500 task.yaml            \u2190 CORAL configuration\n<\/code><\/pre>\n\n\n\n<h4 class=\"wp-block-heading\">Step 2 \u2014 Write the Seed Codebase<\/h4>\n\n\n\n<p>The seed is the starting code that agents will iterate on. It should be a <em>working but naive<\/em> baseline. Agents will improve on it through evolutionary search.<\/p>\n\n\n\n<pre class=\"wp-block-code\"><code># examples\/tsp\/seed\/solution.py\nimport random\n\n# Restate the problem here \u2014 agents cannot read grader.py\nrandom.seed(42)\nCITIES = &#91;(random.random(), random.random()) for _ in range(100)]\n\n# Naive baseline: visit cities in index order\nfor i in range(len(CITIES)):\n    print(i)\n<\/code><\/pre>\n\n\n\n<p><strong>Why seed matters:<\/strong> The better your seed, the less time agents spend on trivial improvements. You can provide an empty seed, but agents will take longer to converge.<\/p>\n\n\n\n<h4 class=\"wp-block-heading\">Step 3 \u2014 Write the Grader (Reward Function)<\/h4>\n\n\n\n<p>This is the most critical component. Subclass <code>TaskGrader<\/code> and implement <code>evaluate()<\/code>. The base class provides two helpers: <code>self.run_program(filename)<\/code> which runs a file from the agent&#8217;s codebase in a subprocess and returns a <code>CompletedProcess<\/code> (with <code>.stdout<\/code>, <code>.stderr<\/code>, <code>.returncode<\/code>), and <code>self.fail(reason)<\/code> which records the failure and returns a null score.<\/p>\n\n\n\n<pre class=\"wp-block-code\"><code># examples\/tsp\/eval\/grader.py\nimport math\nimport random\nfrom coral.grader import TaskGrader, ScoreBundle\n\nrandom.seed(42)\nCITIES = &#91;(random.random(), random.random()) for _ in range(100)]\n\nclass Grader(TaskGrader):\n    def evaluate(self) -&gt; float | ScoreBundle:\n        try:\n            result = self.run_program(\"solution.py\")\n            order = &#91;int(x) for x in result.stdout.strip().split(\"\\\\n\")]\n            assert sorted(order) == list(range(len(CITIES)))\n            dist = sum(\n                math.dist(CITIES&#91;order&#91;i]], CITIES&#91;order&#91;(i + 1) % len(order)]])\n                for i in range(len(order))\n            )\n            return -dist  # shorter = higher score\n        except Exception as e:\n            return self.fail(str(e))\n<\/code><\/pre>\n\n\n\n<p><strong>Grader design principles:<\/strong><\/p>\n\n\n\n<figure class=\"wp-block-table\"><table class=\"has-fixed-layout\"><thead><tr><th>Principle<\/th><th>Explanation<\/th><\/tr><\/thead><tbody><tr><td>Deterministic<\/td><td>Same input must always produce the same score<\/td><\/tr><tr><td>Fast<\/td><td>Each agent submits many attempts; slow graders create bottlenecks<\/td><\/tr><tr><td>No hallucination<\/td><td>Score comes from <em>running actual code<\/em>, not LLM judgment<\/td><\/tr><tr><td>Scalar output<\/td><td>Return a float (higher = better) or use <code>ScoreBundle<\/code> for multi-metric<\/td><\/tr><tr><td>Fail gracefully<\/td><td>Always wrap in try\/except and call <code>self.fail()<\/code><\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n<h4 class=\"wp-block-heading\">Step 4 \u2014 Write the Task Configuration (task.yaml)<\/h4>\n\n\n\n<p>This YAML file is the control center for everything CORAL does.<\/p>\n\n\n\n<pre class=\"wp-block-code\"><code># examples\/tsp\/task.yaml\n\ntask:\n  name: tsp\n  description: |\n    Find the shortest round-trip tour through 100 cities. The coordinates\n    are generated via random with a fixed seed in solution.py.\n    DO NOT MODIFY the seed or CITIES generation!\n\n    solution.py must print 100 integers (0-99) to stdout, one per line,\n    representing the visit order. Each city must appear exactly once.\n\n    The grader computes the total Euclidean round-trip distance\n    and returns -distance as the score (shorter = higher).\n\ngrader:\n  type: function\n  module: eval.grader        # points to eval\/grader.py \u2192 Grader class\n\nagents:\n  count: 1                   # start with 1, scale to 4-8 when ready\n  runtime: claude_code       # or \"codex\" or \"opencode\"\n  model: claude-sonnet-4-6   # use sonnet to test; opus for production\n  max_turns: 200             # agent reboots after this many turns; CORAL keeps running\n\nworkspace:\n  results_dir: \".\/results\"          # where outputs are stored\n  repo_path: \".\/examples\/tsp\/seed\"  # your seed codebase\n<\/code><\/pre>\n\n\n\n<p><strong>Agent count guidance:<\/strong><\/p>\n\n\n\n<figure class=\"wp-block-table\"><table class=\"has-fixed-layout\"><thead><tr><th>Agents<\/th><th>Use Case<\/th><th>Approx. Cost (3h run)<\/th><\/tr><\/thead><tbody><tr><td>1<\/td><td>Testing &amp; debugging<\/td><td>~$15\u201360<\/td><\/tr><tr><td>4<\/td><td>Standard multi-agent run<\/td><td>~$60\u2013240<\/td><\/tr><tr><td>8<\/td><td>Maximum exploration<\/td><td>~$120\u2013480<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n<h4 class=\"wp-block-heading\">Step 5 \u2014 Validate Before Launching<\/h4>\n\n\n\n<p>Always validate your grader before spending API credits:<\/p>\n\n\n\n<pre class=\"wp-block-code\"><code>uv run coral validate examples\/tsp\/task.yaml\n<\/code><\/pre>\n\n\n\n<p>This runs the grader against your seed codebase without spawning any agents. Fix errors here before proceeding.<\/p>\n\n\n\n<h4 class=\"wp-block-heading\">Step 6 \u2014 Launch CORAL<\/h4>\n\n\n\n<pre class=\"wp-block-code\"><code>uv run coral start --config examples\/tsp\/task.yaml\n<\/code><\/pre>\n\n\n\n<p>You should then see CORAL in a tmux session named <code>coral-tsp<\/code>.<\/p>\n\n\n\n<p>CORAL will automatically:<\/p>\n\n\n\n<ol class=\"wp-block-list\">\n<li>Create a git repository from your seed directory<\/li>\n\n\n\n<li>Create isolated git worktrees for each agent<\/li>\n\n\n\n<li>Set up <code>.coral\/public\/<\/code> shared state directory<\/li>\n\n\n\n<li>Symlink the shared state into each worktree<\/li>\n\n\n\n<li>Generate <code>CORAL.md<\/code> instruction files for each agent<\/li>\n\n\n\n<li>Spawn the coding agents as subprocesses<\/li>\n\n\n\n<li>Start the heartbeat manager in the background<\/li>\n<\/ol>\n\n\n\n<h4 class=\"wp-block-heading\">Step 7 \u2014 Monitor Progress<\/h4>\n\n\n\n<pre class=\"wp-block-code\"><code># Open web dashboard (port 8420)\nuv sync --extra ui &amp;&amp; uv run coral ui\n\n# CLI leaderboard (top 20 attempts)\nuv run coral log\n\n# Agent health + real-time status\nuv run coral status\n\n# View notes written by agents\nuv run coral notes\n\n# View abstracted skills discovered by agents\nuv run coral skills\n<\/code><\/pre>\n\n\n\n<h4 class=\"wp-block-heading\">Step 8 \u2014 Override Config at Runtime<\/h4>\n\n\n\n<p>You can override YAML values directly from the CLI without editing files:<\/p>\n\n\n\n<pre class=\"wp-block-code\"><code># Launch with 4 agents instead of 1\nuv run coral start -c examples\/tsp\/task.yaml agents.count=4\n\n# Use a different model\nuv run coral start -c examples\/tsp\/task.yaml agents.model=opus\n<\/code><\/pre>\n\n\n\n<h3 class=\"wp-block-heading\">Part 5 \u2014 Understanding the Shared State Directory<\/h3>\n\n\n\n<p>Shared state (attempts, notes, skills) lives in <code>.coral\/public\/<\/code> and is symlinked into every worktree \u2014 agents see each other&#8217;s work in real time with zero sync overhead.<\/p>\n\n\n\n<pre class=\"wp-block-code\"><code>.coral\/\n\u251c\u2500\u2500 public\/                    \u2190 shared across ALL agents (symlinked)\n\u2502   \u251c\u2500\u2500 attempts\/              \u2190 JSON logs of every eval, keyed by commit hash\n\u2502   \u251c\u2500\u2500 notes\/                 \u2190 Markdown hypotheses written by agents\n\u2502   \u2514\u2500\u2500 skills\/                \u2190 Reusable code modules from successful runs\n\u251c\u2500\u2500 logs\/\n\u2502   \u2514\u2500\u2500 agent_{id}.log         \u2190 Per-agent terminal logs\n\u2514\u2500\u2500 heartbeat\/\n    \u2514\u2500\u2500 config.json            \u2190 Heartbeat trigger configuration\n\nagent_1\/                       \u2190 Agent 1's isolated git worktree\n\u251c\u2500\u2500 .coral \u2192 ..\/.coral\/public  \u2190 symlink to shared state\n\u251c\u2500\u2500 CORAL.md                   \u2190 agent instruction file (auto-generated)\n\u251c\u2500\u2500 solution.py                \u2190 agent's current working code\n\u2514\u2500\u2500 ...\n\nagent_2\/                       \u2190 Agent 2's isolated worktree\n\u2514\u2500\u2500 ...\n<\/code><\/pre>\n\n\n\n<h4 class=\"wp-block-heading\">The Three Artifact Types Explained<\/h4>\n\n\n\n<p><strong>Attempts<\/strong> (<code>attempts\/*.json<\/code>)<\/p>\n\n\n\n<pre class=\"wp-block-code\"><code>{\n  \"commit_hash\": \"a3f92bc\",\n  \"agent_id\": \"agent_1\",\n  \"score\": -45.23,\n  \"description\": \"Applied 2-opt local search after nearest-neighbor init\",\n  \"timestamp\": \"2026-04-26T08:14:22\",\n  \"status\": \"success\"\n}\n<\/code><\/pre>\n\n\n\n<p><strong>Notes<\/strong> (<code>notes\/*.md<\/code>)<\/p>\n\n\n\n<pre class=\"wp-block-code\"><code>---\nagent: agent_2\ntimestamp: 2026-04-26T09:02:11\ntags: &#91;hypothesis, topology]\n---\n# Hypothesis: Or-opt moves may outperform 2-opt on this instance\n\nThe 2-opt implementation converges quickly but gets stuck around -43.x.\nObserved that agent_1's nearest-neighbor init creates long crossing edges\nin the middle cluster. Or-opt with segment size 3 might resolve this.\nNext: try 3-opt with Lin-Kernighan style moves.\n<\/code><\/pre>\n\n\n\n<p><strong>Skills<\/strong> (<code>skills\/<\/code>)<\/p>\n\n\n\n<pre class=\"wp-block-code\"><code>skills\/\n\u2514\u2500\u2500 two_opt_optimizer\/\n    \u251c\u2500\u2500 SKILL.md              \u2190 description, usage, when to apply\n    \u2514\u2500\u2500 two_opt.py            \u2190 reusable implementation\n<\/code><\/pre>\n\n\n\n<h3 class=\"wp-block-heading\">Part 6 \u2014 The CORAL.md Agent Instruction File<\/h3>\n\n\n\n<p>CORAL auto-generates a <code>CORAL.md<\/code> in each agent&#8217;s worktree. This is the agent&#8217;s operating manual \u2014 the deterministic behavioral scaffold that tells the frozen LLM how to work. Key sections include:<\/p>\n\n\n\n<pre class=\"wp-block-code\"><code># CORAL Agent Instructions\n\n## Your Identity\nYou are Agent 1 of 4 working on: TSP (100-city optimization)\n\n## Orientation (do this FIRST, before any code)\n1. Read this task description carefully\n2. Read key files to understand current code state\n3. Check the leaderboard: `coral log` \u2014 find the best current score\n4. Check recent agent activity: what are others trying?\n5. Inspect top attempts by hash: `coral show &lt;hash&gt;`\n6. Search for prior art: keywords relevant to this problem\n7. Read notes from other agents \u2014 what hypotheses exist?\n8. Check available skills in the shared system\n\n## Workflow Loop (repeat continuously)\nPLAN \u2192 EDIT \u2192 EVAL \u2192 REPEAT\n\n### Plan\n- Review what worked. Check coral logs.\n- Inspect top attempts and notes from teammates.\n- Think creatively. What hasn't been tried?\n\n### Edit\n- Modify solution.py (and helper files if needed)\n- Do not modify the seed or evaluation logic\n\n### Eval\n- Run: `coral eval -m \"description of what you tried\"`\n- This stages, commits, and grades in one shot\n- Read the score. Understand why it improved or didn't.\n\n## Ground Rules\n- You are fully autonomous. Do not ask for permission.\n- Write notes frequently: `coral notes add \"your hypothesis\"`\n- Extract skills from successful approaches\n- Collaborate: learn from your agent teammates\n- Never modify grader.py or task configuration\n<\/code><\/pre>\n\n\n\n<h3 class=\"wp-block-heading\">Part 7 \u2014 The Heartbeat Protocol<\/h3>\n\n\n\n<p>The heartbeat is CORAL&#8217;s mechanism for escaping local minima. It runs asynchronously in the background and can interrupt agents with two types of triggers:<\/p>\n\n\n\n<h4 class=\"wp-block-heading\">View and Modify Heartbeat Config<\/h4>\n\n\n\n<pre class=\"wp-block-code\"><code>uv run coral heartbeat          # view current config\nuv run coral heartbeat add      # add a new action\nuv run coral heartbeat remove   # remove an action\n<\/code><\/pre>\n\n\n\n<h4 class=\"wp-block-heading\">Trigger Types<\/h4>\n\n\n\n<p><strong>Interval Trigger<\/strong> \u2014 fires after N evaluation attempts:<\/p>\n\n\n\n<pre class=\"wp-block-code\"><code>{\n  \"type\": \"interval\",\n  \"every_n_evals\": 10,\n  \"action\": \"consolidate_skills\",\n  \"prompt\": \"You have run 10 evaluations. Stop coding. Read all notes in .coral\/public\/notes\/. Synthesize the key findings into a new skill file. What is the best approach discovered so far? What should the next agent prioritize?\"\n}\n<\/code><\/pre>\n\n\n\n<p><strong>Plateau Trigger<\/strong> \u2014 fires when scores stop improving:<\/p>\n\n\n\n<pre class=\"wp-block-code\"><code>{\n  \"type\": \"plateau\",\n  \"patience\": 5,\n  \"action\": \"orthogonal_approach\",\n  \"prompt\": \"Your last 5 attempts have not improved the score. You are in a local minimum. Stop the current approach entirely. Think of a mathematically orthogonal method you have not tried. What would a completely different algorithm look like?\"\n}\n<\/code><\/pre>\n\n\n\n<p>This is conceptually equivalent to <strong>thermal noise injection in simulated annealing<\/strong> \u2014 applying random perturbation to escape gravitational wells in the search space.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">Part 8 \u2014 Multi-Agent Scale-Up<\/h3>\n\n\n\n<p>Once your single-agent run works, scaling to multi-agent is just one config change:<\/p>\n\n\n\n<pre class=\"wp-block-code\"><code>agents:\n  count: 4           # scale from 1 \u2192 4\n  runtime: claude_code\n  model: claude-opus-4-6   # upgrade to opus for production\n<\/code><\/pre>\n\n\n\n<pre class=\"wp-block-code\"><code>uv run coral start -c examples\/tsp\/task.yaml agents.count=4\n<\/code><\/pre>\n\n\n\n<p>While a single autonomous agent can already outperform strong state-of-the-art baselines, a population of agents can push performance substantially further. On Anthropic&#8217;s take-home task for a kernel engineer role, a single agent improved the state of the art from 1,363 cycles to 1,350, while a population of four agents pushed it dramatically further to 1,103.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">Part 9 \u2014 Stop, Resume, and Iterate<\/h3>\n\n\n\n<pre class=\"wp-block-code\"><code># Stop all agents cleanly\nuv run coral stop\n\n# Resume from where you left off (session state is preserved)\nuv run coral resume\n\n# View all past runs\nuv run coral runs\n\n# Inspect a specific attempt by commit hash\nuv run coral show a3f92bc\n\n# Reset a worktree to a previous good state\nuv run coral checkout a3f92bc\n\n# Undo last commit in current worktree\nuv run coral revert\n<\/code><\/pre>\n\n\n\n<h3 class=\"wp-block-heading\">Part 10 \u2014 Complete CLI Reference<\/h3>\n\n\n\n<p>CORAL provides 17+ commands across 5 modules:<\/p>\n\n\n\n<figure class=\"wp-block-table\"><table class=\"has-fixed-layout\"><thead><tr><th>Command<\/th><th>What It Does<\/th><\/tr><\/thead><tbody><tr><td><code>coral init &lt;name&gt;<\/code><\/td><td>Scaffold a new task skeleton<\/td><\/tr><tr><td><code>coral validate &lt;config&gt;<\/code><\/td><td>Test grader against seed, no agents<\/td><\/tr><tr><td><code>coral start -c task.yaml<\/code><\/td><td>Launch all agents<\/td><\/tr><tr><td><code>coral resume<\/code><\/td><td>Resume previous run<\/td><\/tr><tr><td><code>coral stop<\/code><\/td><td>Stop all agents<\/td><\/tr><tr><td><code>coral status<\/code><\/td><td>Agent health + leaderboard<\/td><\/tr><tr><td><code>coral log<\/code><\/td><td>Top 20 attempts leaderboard<\/td><\/tr><tr><td><code>coral log --recent<\/code><\/td><td>Most recent attempts<\/td><\/tr><tr><td><code>coral log --search \"query\"<\/code><\/td><td>Search attempts by description<\/td><\/tr><tr><td><code>coral show &lt;hash&gt;<\/code><\/td><td>Full attempt details + code diff<\/td><\/tr><tr><td><code>coral notes<\/code><\/td><td>Browse all agent notes<\/td><\/tr><tr><td><code>coral skills<\/code><\/td><td>Browse all abstracted skills<\/td><\/tr><tr><td><code>coral runs<\/code><\/td><td>List all historical runs<\/td><\/tr><tr><td><code>coral ui<\/code><\/td><td>Web dashboard (port 8420)<\/td><\/tr><tr><td><code>coral eval -m \"desc\"<\/code><\/td><td>Stage + commit + grade (used by agents)<\/td><\/tr><tr><td><code>coral diff<\/code><\/td><td>Show uncommitted changes<\/td><\/tr><tr><td><code>coral revert<\/code><\/td><td>Undo last commit<\/td><\/tr><tr><td><code>coral checkout &lt;hash&gt;<\/code><\/td><td>Reset to previous attempt<\/td><\/tr><tr><td><code>coral heartbeat<\/code><\/td><td>View\/modify heartbeat config<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n<h3 class=\"wp-block-heading\">Part 11 \u2014 Available Example Tasks<\/h3>\n\n\n\n<p>Ready-to-run task configurations are provided in <code>examples\/<\/code>:<\/p>\n\n\n\n<figure class=\"wp-block-table\"><table class=\"has-fixed-layout\"><thead><tr><th>Task<\/th><th>Domain<\/th><th>What Agents Optimize<\/th><\/tr><\/thead><tbody><tr><td><code>circle_packing<\/code><\/td><td>Optimization<\/td><td>Pack 26 circles into unit square<\/td><\/tr><tr><td><code>erdos<\/code><\/td><td>Mathematics<\/td><td>Solve Erd\u0151s Min Overlap conjecture<\/td><\/tr><tr><td><code>kernel_builder<\/code><\/td><td>Systems<\/td><td>VLIW SIMD kernel performance<\/td><\/tr><tr><td><code>kernel_engineering<\/code><\/td><td>Systems<\/td><td>GPU kernel cycle count<\/td><\/tr><tr><td><code>mnist<\/code><\/td><td>Machine Learning<\/td><td>Digit classification accuracy<\/td><\/tr><tr><td><code>spaceship_titanic<\/code><\/td><td>ML\/Kaggle<\/td><td>Passenger survival prediction<\/td><\/tr><tr><td><code>stanford_covid_vaccine<\/code><\/td><td>Bio\/ML<\/td><td>mRNA degradation rate<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n<p>Run any example directly:<\/p>\n\n\n\n<pre class=\"wp-block-code\"><code>uv run coral start -c examples\/circle_packing\/task.yaml\nuv run coral start -c examples\/kernel_engineering\/task.yaml\n<\/code><\/pre>\n\n\n\n<h3 class=\"wp-block-heading\">Part 12 \u2014 Key Design Decisions &amp; Warnings<\/h3>\n\n\n\n<p><strong>What CORAL updates vs. what it doesn&#8217;t:<\/strong><\/p>\n\n\n\n<figure class=\"wp-block-table\"><table class=\"has-fixed-layout\"><thead><tr><th>Updated During a Run<\/th><th>Never Updated<\/th><\/tr><\/thead><tbody><tr><td><code>.coral\/public\/attempts\/*.json<\/code><\/td><td>LLM model weights<\/td><\/tr><tr><td><code>.coral\/public\/notes\/*.md<\/code><\/td><td>Grader logic<\/td><\/tr><tr><td><code>.coral\/public\/skills\/<\/code><\/td><td>Task configuration<\/td><\/tr><tr><td>Agent worktree code files<\/td><td>Token context window of any model<\/td><\/tr><tr><td>Heartbeat configuration<\/td><td><\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n<p><strong>Cost Management \u2014 Critical:<\/strong><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>Set hard API spending limits in your provider dashboard <em>before<\/em> launching<\/li>\n\n\n\n<li>Start with <code>agents.count=1<\/code> and <code>model: sonnet<\/code> for testing<\/li>\n\n\n\n<li>Upgrade to <code>count=4<\/code> and <code>model: opus<\/code> only for production runs<\/li>\n\n\n\n<li>A 12-hour, 8-agent Opus run can cost <strong>$500+<\/strong> if uncapped<\/li>\n<\/ul>\n\n\n\n<p><strong>Budget-aware launch pattern:<\/strong><\/p>\n\n\n\n<pre class=\"wp-block-code\"><code># Test run: 1 agent, sonnet, 50 turns max\nuv run coral start -c task.yaml agents.count=1 agents.model=claude-sonnet-4-6 agents.max_turns=50\n\n# Production run: 4 agents, opus, unlimited turns (stop manually)\nuv run coral start -c task.yaml agents.count=4 agents.model=claude-opus-4-6\n<\/code><\/pre>\n\n\n\n<h3 class=\"wp-block-heading\">Summary: The CORAL Setup Checklist<\/h3>\n\n\n\n<pre class=\"wp-block-code\"><code>\u2705 1. Install uv\n\u2705 2. Install &amp; authenticate your coding agent (Claude Code \/ Codex \/ OpenCode)\n\u2705 3. Clone CORAL repo and run `uv sync`\n\u2705 4. Create task directory: seed\/ + eval\/\n\u2705 5. Write seed codebase (naive working baseline)\n\u2705 6. Write grader.py (subclass TaskGrader, implement evaluate())\n\u2705 7. Write task.yaml (link seed, grader, agent config)\n\u2705 8. Validate: `uv run coral validate task.yaml`\n\u2705 9. Set API spending limits!\n\u2705 10. Launch: `uv run coral start -c task.yaml`\n\u2705 11. Monitor: `uv run coral ui` or `uv run coral status`\n\u2705 12. Iterate: stop \u2192 adjust heartbeat \u2192 resume\n<\/code><\/pre>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"video\">\ud83c\udfacVideo about CORAL<\/h2>\n\n\n\n<figure class=\"wp-block-embed is-type-video is-provider-youtube wp-block-embed-youtube wp-embed-aspect-16-9 wp-has-aspect-ratio\"><div class=\"wp-block-embed__wrapper\">\n<iframe loading=\"lazy\" title=\"SuperIntelligence: Why the Future of AI is a File System (CORAL)\" width=\"500\" height=\"281\" src=\"https:\/\/www.youtube.com\/embed\/7n5EVMtYA4I?feature=oembed\" frameborder=\"0\" allow=\"accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture; web-share\" referrerpolicy=\"strict-origin-when-cross-origin\" allowfullscreen><\/iframe>\n<\/div><\/figure>\n\n\n\n<div class=\"wp-block-group has-pale-cyan-blue-background-color has-background\"><div class=\"wp-block-group__inner-container is-layout-constrained wp-block-group-is-layout-constrained\">\n<h3 class=\"wp-block-heading\">\ud83d\udd0d Related Concepts and Context in the Video<\/h3>\n\n\n\n<p><strong>Why not train the LLM?<\/strong> The presenter&#8217;s central critique \u2014 and genuine puzzle \u2014 is that elite research institutions are deliberately <em>not<\/em> updating model weights, even when new knowledge is being discovered during runs. The argument is economic and practical: file-system-based intelligence is cheaper, debuggable, human-readable, and deployable without retraining cycles.<\/p>\n\n\n\n<p><strong>The ADI Framing<\/strong> The presenter reframes this class of systems as &#8220;Advanced Intelligence of the Environment&#8221; (ADI), contrasting it with AGI. The suggestion is that what institutions are building is not a smarter <em>mind<\/em>, but a smarter <em>world<\/em> around a frozen mind.<\/p>\n\n\n\n<p><strong>Connections to Prior Work<\/strong> The presenter references <strong>OpenClaw<\/strong> and <strong>AutoResearch-Claw<\/strong> as conceptual predecessors \u2014 autonomous agent loops with no human interaction that spawn agents, share knowledge, and iterate. Early autonomous agent frameworks like AutoGPT and BabyAGI established the pattern of AI programs that can think for themselves, create tasks, and reprioritize their task list to achieve a given objective \u2014 CORAL extends this into a far more structured, multi-agent, file-system-native paradigm.<\/p>\n\n\n\n<p><strong>Open-Ended Complexity<\/strong> Unlike benchmark-oriented systems, CORAL targets <strong>open-ended problem spaces<\/strong> \u2014 mathematical optimization, theoretical physics conjectures, and similar domains where the solution topology is unknown. The file system becomes the mechanism for agents to build cumulative knowledge maps across potentially days-long runs.<\/p>\n\n\n\n<p><strong>Cost Warning<\/strong> The presenter issues a practical caution: a single 3-hour, 4-agent CORAL run using Claude Opus 4 can cost up to <strong>$240 USD<\/strong>. Extended overnight multi-agent runs could produce significant unexpected API charges without proper credit limits in place.<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n<\/div><\/div>\n\n\n\n<h2 class=\"wp-block-heading\">\u2705 Conclusion and Key Takeaways<\/h2>\n\n\n\n<p>CORAL represents a meaningful inflection point in AI systems design \u2014 not because it makes models smarter, but because it makes the <em>context around models<\/em> dramatically richer, more persistent, and more collaborative. Whether this is the &#8220;right&#8221; path forward is a question the presenter leaves deliberately open. The frozen-model, file-system-native approach trades the unpredictability of fine-tuning for the predictability of deterministic markdown templates \u2014 but also carries genuine risks: smaller LLMs hallucinating during complex synthesis tasks, costs scaling rapidly with agent count, and a system architecture that depends heavily on the linguistic reasoning ability of the very LLM it refuses to update.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\"><strong>Key Takeaways:<\/strong><\/h3>\n\n\n\n<ul class=\"wp-block-list\">\n<li>CORAL is an autonomous multi-agent infrastructure from MIT\/Stanford\/Meta \u2014 not just a framework but a full operational system<\/li>\n\n\n\n<li>Intelligence in CORAL lives in the <strong>file system harness<\/strong>, not the model \u2014 all LLMs remain frozen<\/li>\n\n\n\n<li>Three artifact types (Attempts, Notes, Skills) form the shared memory layer across all agents<\/li>\n\n\n\n<li>The <strong>Heartbeat Protocol<\/strong> is CORAL&#8217;s mechanism for escaping local minima \u2014 analogous to simulated annealing<\/li>\n\n\n\n<li>This is <strong>gradient-free, inference-time<\/strong> intelligence \u2014 no training, no fine-tuning, no weight updates<\/li>\n\n\n\n<li><strong>Cost discipline is critical<\/strong>: multi-agent runs with frontier models can become very expensive very quickly<\/li>\n\n\n\n<li>The presenter coins <strong>ADI (Advanced Intelligence)<\/strong> to distinguish this paradigm from classical AI\/AGI framing<\/li>\n\n\n\n<li>The GitHub repo is public, MIT licensed, and actively maintained: <a href=\"http:\/\/github.com\/corl-team\">github.com\/corl-team<\/a><\/li>\n<\/ul>\n\n\n\n<h3 class=\"wp-block-heading\">\ud83d\udcda Related References<\/h3>\n\n\n\n<ul class=\"wp-block-list\">\n<li>GitHub: <a href=\"https:\/\/github.com\/Human-Agent-Society\/CORAL\">https:\/\/github.com\/Human-Agent-Society\/CORAL<\/a><\/li>\n\n\n\n<li>Docs: <a href=\"https:\/\/docs.coralxyz.com\">https:\/\/docs.coralxyz.com<\/a><\/li>\n\n\n\n<li>Blog\/Paper: <a href=\"https:\/\/human-agent-society.github.io\/CORAL\/\">https:\/\/human-agent-society.github.io\/CORAL\/<\/a><\/li>\n\n\n\n<li>ArXiv Paper: <a href=\"https:\/\/arxiv.org\/abs\/2604.01658\">https:\/\/arxiv.org\/abs\/2604.01658<\/a><\/li>\n<\/ul>\n","protected":false},"excerpt":{"rendered":"<p>CORAL is a multi-agent infrastructure where autonomous AI agents continuously explore, evaluate, and improve solutions \u2014 not by retraining the model, but by writing to a shared file system of attempts, notes, and skills. Built by MIT, Stanford, and Meta, it runs frozen LLMs inside isolated git worktrees, coordinated through persistent shared memory and a heartbeat protocol that prevents stagnation \u2014 intelligence engineered around the model, not inside it.<\/p>\n","protected":false},"author":1,"featured_media":8307,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[15,18,13,7],"tags":[],"class_list":["post-8305","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-ai","category-education","category-quantum-and-u","category-quantum-mindset-programme"],"aioseo_notices":[],"featured_image_src":"https:\/\/meta-quantum.today\/wp-content\/uploads\/2026\/04\/SuperIntelligence-Why-the-Future-of-AI-is-a-File-System-CORAL.jpg","featured_image_src_square":"https:\/\/meta-quantum.today\/wp-content\/uploads\/2026\/04\/SuperIntelligence-Why-the-Future-of-AI-is-a-File-System-CORAL.jpg","author_info":{"display_name":"coffee","author_link":"https:\/\/meta-quantum.today\/?author=1"},"rbea_author_info":{"display_name":"coffee","author_link":"https:\/\/meta-quantum.today\/?author=1"},"rbea_excerpt_info":"CORAL is a multi-agent infrastructure where autonomous AI agents continuously explore, evaluate, and improve solutions \u2014 not by retraining the model, but by writing to a shared file system of attempts, notes, and skills. Built by MIT, Stanford, and Meta, it runs frozen LLMs inside isolated git worktrees, coordinated through persistent shared memory and a heartbeat protocol that prevents stagnation \u2014 intelligence engineered around the model, not inside it.","category_list":"<a href=\"https:\/\/meta-quantum.today\/?cat=15\" rel=\"category\">AI<\/a>, <a href=\"https:\/\/meta-quantum.today\/?cat=18\" rel=\"category\">Education<\/a>, <a href=\"https:\/\/meta-quantum.today\/?cat=13\" rel=\"category\">Quantum and U<\/a>, <a href=\"https:\/\/meta-quantum.today\/?cat=7\" rel=\"category\">Quantum Mindset Programme<\/a>","comments_num":"0 comments","_links":{"self":[{"href":"https:\/\/meta-quantum.today\/index.php?rest_route=\/wp\/v2\/posts\/8305","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/meta-quantum.today\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/meta-quantum.today\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/meta-quantum.today\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/meta-quantum.today\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=8305"}],"version-history":[{"count":5,"href":"https:\/\/meta-quantum.today\/index.php?rest_route=\/wp\/v2\/posts\/8305\/revisions"}],"predecessor-version":[{"id":8312,"href":"https:\/\/meta-quantum.today\/index.php?rest_route=\/wp\/v2\/posts\/8305\/revisions\/8312"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/meta-quantum.today\/index.php?rest_route=\/wp\/v2\/media\/8307"}],"wp:attachment":[{"href":"https:\/\/meta-quantum.today\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=8305"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/meta-quantum.today\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=8305"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/meta-quantum.today\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=8305"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}