feat: Imitate tab — pull CF product samples, compare LLM responses

Backend (app/imitate.py):
- GET /api/imitate/products — reads imitate: config, checks online status
- GET /api/imitate/products/{id}/sample — fetches real item from product API
- GET /api/imitate/run (SSE) — streams ollama responses for selected models
- POST /api/imitate/push-corrections — queues results in SFT corrections JSONL

Frontend (ImitateView.vue):
- Step 1: product picker grid (online/offline status, icon from config)
- Step 2: raw sample preview + editable prompt textarea
- Step 3: ollama model multi-select, temperature slider, SSE run with live log
- Step 4: response cards side by side, push to Corrections button

Wiring:
- app/api.py: include imitate_router at /api/imitate
- web/src/router: /imitate route + lazy import
- AppSidebar: Imitate nav entry (mirror icon)
- config/label_tool.yaml.example: imitate: section with peregrine example
- 16 unit tests (100% passing)

Also: BenchmarkView.vue Compare panel — side-by-side run diff for bench results

.env.example

@@ -0,0 +1,19 @@
+# Avocet — environment variable configuration
+# Copy to .env and fill in values. All keys are optional.
+# label_tool.yaml takes precedence over env vars where both exist.
+
+# ── Local inference (Ollama) ───────────────────────────────────────────────────
+# OLLAMA_HOST defaults to http://localhost:11434 if unset.
+OLLAMA_HOST=http://localhost:11434
+OLLAMA_MODEL=llama3.2:3b
+
+# ── cf-orch coordinator (paid/premium tiers) ───────────────────────────────────
+# Required for multi-GPU LLM benchmarking via the cf-orch benchmark harness.
+# Free-tier users can leave these unset and use Ollama only.
+CF_ORCH_URL=http://localhost:7700
+CF_LICENSE_KEY=CFG-AVCT-xxxx-xxxx-xxxx
+
+# ── Cloud LLM backends (optional — paid/premium) ──────────────────────────────
+# Set one of these to use a cloud LLM instead of a local model.
+# ANTHROPIC_API_KEY=sk-ant-...
+# OPENAI_API_KEY=sk-...

app/cforch.py

@@ -14,6 +14,7 @@ from __future__ import annotations
 
 import json
 import logging
+import os
 import re
 import subprocess as _subprocess
 from pathlib import Path
@@ -49,16 +50,32 @@ def _config_file() -> Path:
 
 
 def _load_cforch_config() -> dict:
-    """Read label_tool.yaml and return the cforch sub-dict (or {} if absent/malformed)."""
+    """Read label_tool.yaml cforch section, falling back to environment variables.
+
+    Priority (highest to lowest):
+      1. label_tool.yaml cforch: key
+      2. Environment variables (CF_ORCH_URL, CF_LICENSE_KEY, OLLAMA_HOST, OLLAMA_MODEL)
+    """
     f = _config_file()
-    if not f.exists():
-        return {}
-    try:
-        raw = yaml.safe_load(f.read_text(encoding="utf-8")) or {}
-    except yaml.YAMLError as exc:
-        logger.warning("Failed to parse cforch config %s: %s", f, exc)
-        return {}
-    return raw.get("cforch", {}) or {}
+    file_cfg: dict = {}
+    if f.exists():
+        try:
+            raw = yaml.safe_load(f.read_text(encoding="utf-8")) or {}
+            file_cfg = raw.get("cforch", {}) or {}
+        except yaml.YAMLError as exc:
+            logger.warning("Failed to parse cforch config %s: %s", f, exc)
+
+    # Env var fallbacks — only used when the yaml key is absent or empty
+    def _coalesce(file_val: str, env_key: str) -> str:
+        return file_val if file_val else os.environ.get(env_key, "")
+
+    return {
+        **file_cfg,
+        "coordinator_url": _coalesce(file_cfg.get("coordinator_url", ""), "CF_ORCH_URL"),
+        "license_key":     _coalesce(file_cfg.get("license_key", ""),     "CF_LICENSE_KEY"),
+        "ollama_url":      _coalesce(file_cfg.get("ollama_url", ""),       "OLLAMA_HOST"),
+        "ollama_model":    _coalesce(file_cfg.get("ollama_model", ""),     "OLLAMA_MODEL"),
+    }
 
 
 def _strip_ansi(text: str) -> str:
@@ -186,7 +203,8 @@ def run_benchmark(
     results_dir = cfg.get("results_dir", "")
     python_bin = cfg.get("python_bin", "/devl/miniconda3/envs/cf/bin/python")
     cfg_coordinator = cfg.get("coordinator_url", "")
-    cfg_ollama = cfg.get("ollama_url", "")
+    cfg_ollama      = cfg.get("ollama_url", "")
+    cfg_license_key = cfg.get("license_key", "")
 
     def generate():
         global _BENCH_RUNNING, _bench_proc
@@ -208,13 +226,19 @@ def run_benchmark(
         if model_tags:
             cmd.extend(["--filter-tags"] + model_tags.split(","))
 
+        # query param overrides config, config overrides env var (already resolved by _load_cforch_config)
         effective_coordinator = coordinator_url if coordinator_url else cfg_coordinator
-        effective_ollama = ollama_url if ollama_url else cfg_ollama
+        effective_ollama      = ollama_url      if ollama_url      else cfg_ollama
         if effective_coordinator:
             cmd.extend(["--coordinator", effective_coordinator])
         if effective_ollama:
             cmd.extend(["--ollama-url", effective_ollama])
 
+        # Pass license key as env var so subprocess can authenticate with cf-orch
+        proc_env = {**os.environ}
+        if cfg_license_key:
+            proc_env["CF_LICENSE_KEY"] = cfg_license_key
+
         _BENCH_RUNNING = True
         try:
             proc = _subprocess.Popen(
@@ -223,6 +247,7 @@ def run_benchmark(
                 stderr=_subprocess.STDOUT,
                 text=True,
                 bufsize=1,
+                env=proc_env,
             )
             _bench_proc = proc
             try:
@@ -256,6 +281,25 @@ def run_benchmark(
     )
 
 
+# ── GET /config ────────────────────────────────────────────────────────────────
+
+@router.get("/config")
+def get_cforch_config() -> dict:
+    """Return resolved cf-orch connection config (env vars merged with yaml).
+
+    Redacts license_key — only returns whether it is set, not the value.
+    Used by the Settings UI to show current connection state.
+    """
+    cfg = _load_cforch_config()
+    return {
+        "coordinator_url": cfg.get("coordinator_url", ""),
+        "ollama_url":      cfg.get("ollama_url", ""),
+        "ollama_model":    cfg.get("ollama_model", ""),
+        "license_key_set": bool(cfg.get("license_key", "")),
+        "source": "env" if not _config_file().exists() else "yaml+env",
+    }
+
+
 # ── GET /results ───────────────────────────────────────────────────────────────
 
 @router.get("/results")

app/sft.py

@@ -51,17 +51,26 @@ def _config_file() -> Path:
     return _ROOT / "config" / "label_tool.yaml"
 
 
+_DEFAULT_BENCH_RESULTS_DIR = "/Library/Development/CircuitForge/circuitforge-orch/scripts/bench_results"
+
+
+def set_default_bench_results_dir(path: str) -> None:
+    """Override the default bench_results_dir — used by tests to avoid real filesystem."""
+    global _DEFAULT_BENCH_RESULTS_DIR
+    _DEFAULT_BENCH_RESULTS_DIR = path
+
+
 def _get_bench_results_dir() -> Path:
     f = _config_file()
-    if not f.exists():
-        return Path("/nonexistent-bench-results")
-    try:
-        raw = yaml.safe_load(f.read_text(encoding="utf-8")) or {}
-    except yaml.YAMLError as exc:
-        logger.warning("Failed to parse SFT config %s: %s", f, exc)
-        return Path("/nonexistent-bench-results")
-    d = raw.get("sft", {}).get("bench_results_dir", "")
-    return Path(d) if d else Path("/nonexistent-bench-results")
+    if f.exists():
+        try:
+            raw = yaml.safe_load(f.read_text(encoding="utf-8")) or {}
+            d = raw.get("sft", {}).get("bench_results_dir", "")
+            if d:
+                return Path(d)
+        except yaml.YAMLError as exc:
+            logger.warning("Failed to parse SFT config %s: %s", f, exc)
+    return Path(_DEFAULT_BENCH_RESULTS_DIR)
 
 
 def _candidates_file() -> Path:

config/label_tool.yaml.example

@@ -27,6 +27,26 @@ max_per_account: 500
 sft:
   bench_results_dir: /path/to/circuitforge-orch/scripts/bench_results
 
+# cf-orch integration — LLM benchmark harness via cf-orch coordinator.
+# All keys here override the corresponding environment variables.
+# Omit any key to fall back to the env var (see .env.example).
+cforch:
+  # Path to cf-orch's benchmark.py script
+  bench_script: /path/to/circuitforge-orch/scripts/benchmark.py
+  # Task and model definition files (yaml)
+  bench_tasks:  /path/to/circuitforge-orch/scripts/bench_tasks.yaml
+  bench_models: /path/to/circuitforge-orch/scripts/bench_models.yaml
+  # Where benchmark results are written (also used for SFT candidate discovery)
+  results_dir:  /path/to/circuitforge-orch/scripts/bench_results
+  # Python interpreter with cf-orch installed
+  python_bin:   /devl/miniconda3/envs/cf/bin/python
+
+  # Connection config — override env vars CF_ORCH_URL / CF_LICENSE_KEY / OLLAMA_HOST
+  # coordinator_url: http://localhost:7700
+  # license_key:     CFG-AVCT-xxxx-xxxx-xxxx
+  # ollama_url:      http://localhost:11434
+  # ollama_model:    llama3.2:3b
+
 # Imitate tab — pull real samples from sibling CF product APIs and run them
 # through local LLMs to build a corrections dataset.
 # ollama_url defaults to cforch.ollama_url if omitted here.

environment.yml

@@ -22,5 +22,8 @@ dependencies:
     # Optional: BGE reranker adapter
     # - FlagEmbedding
 
+    # CircuitForge shared core (LLM router, tier system, config)
+    - circuitforge-core>=0.9.0
+
     # Dev
     - pytest>=8.0

tests/test_cforch.py

@@ -283,3 +283,87 @@ def test_cancel_terminates_running_benchmark(client):
     mock_proc.terminate.assert_called_once()
     assert cforch_module._BENCH_RUNNING is False
     assert cforch_module._bench_proc is None
+
+
+# ── GET /config ────────────────────────────────────────────────────────────────
+
+def test_config_returns_empty_when_no_yaml_no_env(client, monkeypatch):
+    """No yaml, no env vars — all fields empty, license_key_set False."""
+    for key in ("CF_ORCH_URL", "CF_LICENSE_KEY", "OLLAMA_HOST", "OLLAMA_MODEL"):
+        monkeypatch.delenv(key, raising=False)
+
+    r = client.get("/api/cforch/config")
+    assert r.status_code == 200
+    data = r.json()
+    assert data["coordinator_url"] == ""
+    assert data["ollama_url"] == ""
+    assert data["license_key_set"] is False
+
+
+def test_config_reads_env_vars_when_no_yaml(client, monkeypatch):
+    """Env vars populate fields when label_tool.yaml has no cforch section."""
+    monkeypatch.setenv("CF_ORCH_URL",      "http://orch.example.com:7700")
+    monkeypatch.setenv("CF_LICENSE_KEY",   "CFG-AVCT-TEST-TEST-TEST")
+    monkeypatch.setenv("OLLAMA_HOST",      "http://ollama.local:11434")
+    monkeypatch.setenv("OLLAMA_MODEL",     "mistral:7b")
+
+    r = client.get("/api/cforch/config")
+    assert r.status_code == 200
+    data = r.json()
+    assert data["coordinator_url"] == "http://orch.example.com:7700"
+    assert data["ollama_url"]      == "http://ollama.local:11434"
+    assert data["ollama_model"]    == "mistral:7b"
+    assert data["license_key_set"] is True   # set, but value not exposed
+
+
+def test_config_yaml_overrides_env(client, config_dir, monkeypatch):
+    """label_tool.yaml cforch values take priority over env vars."""
+    monkeypatch.setenv("CF_ORCH_URL",  "http://env-orch:7700")
+    monkeypatch.setenv("OLLAMA_HOST",  "http://env-ollama:11434")
+
+    _write_config(config_dir, {
+        "coordinator_url": "http://yaml-orch:7700",
+        "ollama_url":      "http://yaml-ollama:11434",
+    })
+
+    r = client.get("/api/cforch/config")
+    assert r.status_code == 200
+    data = r.json()
+    assert data["coordinator_url"] == "http://yaml-orch:7700"
+    assert data["ollama_url"]      == "http://yaml-ollama:11434"
+    assert data["source"] == "yaml+env"
+
+
+def test_run_passes_license_key_env_to_subprocess(client, config_dir, tmp_path, monkeypatch):
+    """CF_LICENSE_KEY must be forwarded to the benchmark subprocess env."""
+    monkeypatch.setenv("CF_LICENSE_KEY", "CFG-AVCT-ENV-ONLY-KEY")
+
+    bench_script = tmp_path / "benchmark.py"
+    bench_script.write_text("# stub", encoding="utf-8")
+    tasks_file   = tmp_path / "bench_tasks.yaml"
+    tasks_file.write_text(yaml.dump({"tasks": []}), encoding="utf-8")
+    models_file  = tmp_path / "bench_models.yaml"
+    models_file.write_text(yaml.dump({"models": []}), encoding="utf-8")
+
+    _write_config(config_dir, {
+        "bench_script":  str(bench_script),
+        "bench_tasks":   str(tasks_file),
+        "bench_models":  str(models_file),
+        "results_dir":   str(tmp_path / "results"),
+        "python_bin":    "/usr/bin/python3",
+    })
+
+    captured_env: dict = {}
+
+    def fake_popen(cmd, **kwargs):
+        captured_env.update(kwargs.get("env", {}))
+        mock = MagicMock()
+        mock.stdout = iter([])
+        mock.returncode = 0
+        mock.wait = MagicMock()
+        return mock
+
+    with patch("app.cforch._subprocess.Popen", side_effect=fake_popen):
+        client.get("/api/cforch/run")
+
+    assert captured_env.get("CF_LICENSE_KEY") == "CFG-AVCT-ENV-ONLY-KEY"

tests/test_sft.py

@@ -8,13 +8,16 @@ from pathlib import Path
 @pytest.fixture(autouse=True)
 def reset_sft_globals(tmp_path):
     from app import sft as sft_module
-    _prev_data = sft_module._SFT_DATA_DIR
-    _prev_cfg = sft_module._SFT_CONFIG_DIR
+    _prev_data    = sft_module._SFT_DATA_DIR
+    _prev_cfg     = sft_module._SFT_CONFIG_DIR
+    _prev_default = sft_module._DEFAULT_BENCH_RESULTS_DIR
     sft_module.set_sft_data_dir(tmp_path)
     sft_module.set_sft_config_dir(tmp_path)
+    sft_module.set_default_bench_results_dir(str(tmp_path / "bench_results"))
     yield
     sft_module.set_sft_data_dir(_prev_data)
     sft_module.set_sft_config_dir(_prev_cfg)
+    sft_module.set_default_bench_results_dir(_prev_default)
 
 
 @pytest.fixture

web/src/views/SettingsView.vue

@@ -115,8 +115,18 @@
       <h2 class="section-title">cf-orch Integration</h2>
       <p class="section-desc">
         Import SFT (supervised fine-tuning) candidates from cf-orch benchmark runs.
+        Connection settings fall back to environment variables
+        (<code>CF_ORCH_URL</code>, <code>CF_LICENSE_KEY</code>, <code>OLLAMA_HOST</code>)
+        when not set here.
       </p>
 
+      <!-- Connection status pill -->
+      <div v-if="orchConfig" class="orch-status-row">
+        <span class="orch-status-pill" :class="orchStatusClass">{{ orchStatusLabel }}</span>
+        <span v-if="orchConfig.source === 'env'" class="orch-source-note">via env vars</span>
+        <span v-else class="orch-source-note">via label_tool.yaml</span>
+      </div>
+
       <div class="field-row">
         <label class="field field-grow">
           <span>bench_results_dir</span>
@@ -181,7 +191,7 @@
 </template>
 
 <script setup lang="ts">
-import { ref, onMounted } from 'vue'
+import { ref, computed, onMounted } from 'vue'
 import { useApiFetch } from '../composables/useApi'
 
 interface Account {
@@ -199,12 +209,27 @@ const saveOk        = ref(true)
 const richMotion    = ref(localStorage.getItem('cf-avocet-rich-motion') !== 'false')
 const keyHints      = ref(localStorage.getItem('cf-avocet-key-hints') !== 'false')
 
-// SFT integration state
+// SFT / cf-orch integration state
 const benchResultsDir = ref('')
 const runs            = ref<Array<{ run_id: string; timestamp: string; candidate_count: number; already_imported: boolean }>>([])
 const importingRunId  = ref<string | null>(null)
 const importResult    = ref<{ imported: number; skipped: number } | null>(null)
 const saveStatus      = ref('')
+const orchConfig      = ref<{ coordinator_url: string; ollama_url: string; ollama_model: string; license_key_set: boolean; source: string } | null>(null)
+
+const orchStatusClass = computed(() => {
+  if (!orchConfig.value) return 'status-unknown'
+  if (orchConfig.value.coordinator_url) return 'status-connected'
+  if (orchConfig.value.ollama_url) return 'status-local'
+  return 'status-unconfigured'
+})
+
+const orchStatusLabel = computed(() => {
+  if (!orchConfig.value) return 'Unknown'
+  if (orchConfig.value.coordinator_url) return '● cf-orch coordinator'
+  if (orchConfig.value.ollama_url) return '● Ollama (local)'
+  return '○ Not configured'
+})
 
 async function loadSftConfig() {
   try {
@@ -218,6 +243,15 @@ async function loadSftConfig() {
   }
 }
 
+async function loadOrchConfig() {
+  try {
+    const res = await fetch('/api/cforch/config')
+    if (res.ok) orchConfig.value = await res.json()
+  } catch {
+    // non-fatal
+  }
+}
+
 async function saveSftConfig() {
   saveStatus.value = 'Saving…'
   try {
@@ -337,6 +371,7 @@ function onKeyHintsChange() {
 onMounted(() => {
   reload()
   loadSftConfig()
+  loadOrchConfig()
 })
 </script>
 
@@ -564,6 +599,31 @@ onMounted(() => {
   width: 100%;
 }
 
+.orch-status-row {
+  display: flex;
+  align-items: center;
+  gap: var(--space-2);
+  margin-bottom: var(--space-3);
+}
+
+.orch-status-pill {
+  font-size: 0.8rem;
+  font-weight: 600;
+  padding: var(--space-1) var(--space-3);
+  border-radius: var(--radius-full);
+}
+
+.status-connected    { background: color-mix(in srgb, var(--color-success, #3a7a32) 12%, transparent); color: var(--color-success, #3a7a32); }
+.status-local        { background: color-mix(in srgb, var(--color-primary) 12%, transparent); color: var(--color-primary); }
+.status-unconfigured { background: var(--color-surface-alt); color: var(--color-text-muted); }
+.status-unknown      { background: var(--color-surface-alt); color: var(--color-text-muted); }
+
+.orch-source-note {
+  font-size: 0.75rem;
+  color: var(--color-text-muted);
+  font-style: italic;
+}
+
 .runs-table {
   width: 100%;
   border-collapse: collapse;