chnjinlei/hive
1
0
Fork 0
Code Issues Pull Requests Actions 41 Packages Projects Releases Wiki Activity

feat(quickstart): add Local (Ollama) LLM provider option (#6028)

Browse Source 
* feat(quickstart): add Local (Ollama) LLM provider option
- Detect Ollama via 'ollama list' in quickstart.sh and quickstart.ps1
- Add 'Local (Ollama)' menu option with interactive model picker
- Save provider=ollama, model=<selected> to ~/.hive/configuration.json
- Omit api_key_env_var for Ollama (no API key required)
Refs #5154, #5231

* feat: add local Ollama support and resolve native tool calling

This integrates Ollama as a first-class local provider choice during quickstart, and patches several configuration barriers preventing local models from safely executing the framework's agent graphs.

* **Quickstart Integration**: Added `Local (Ollama)` to the provider menu in both quickstart.sh and quickstart.ps1. When selected, it automatically queries `ollama list` and allows the user to pick an installed model without prompting for an API key.
* **Routing & Configuration**: Automatically sets `"api_base": "http://localhost:11434"` so LiteLLM routes correctly to the local daemon, and increases the default max_tokens config.py allocation to `32768`.
* **Native Tool Calling**: Normalized Ollama models to strictly use the ollama_chat provider prefix inside litellm.py and registered them as `supports_function_calling: True`. This forces native structured function calling and fixes the infinite loop caused by JSON-mode text fallbacks.
* **Context Truncation Fix**: Updated config.py to explicitly pass `"num_ctx": 16384` to Ollama. This prevents the local daemon from silently truncating the Queen agent's ~9,500 token system prompt (Ollama defaults to 2048 `num_ctx`).
* **UX Warnings**: Added terminal notices warning users to select high-parameter models (e.g., `qwen2.5:72b+`) to ensure sufficient contextual reasoning abilities.

Resolves #6027
Resolves #6028

* test: add unit tests for Ollama helper functions

Cover _is_ollama_model(), _ensure_ollama_chat_prefix(), and num_ctx
injection in get_llm_extra_kwargs() as requested in PR review.
Fix existing test_init_ollama_no_key_needed assertion to expect the
normalised ollama_chat/ prefix.

Made-with: Cursor

* chores: fixed merge conflict

* fix(ollama): address PR review comments and normalize provider config

* fix(ollama): align quickstart defaults and add tool_choice comment

* fix(ollama): enforce OLLAMA_DETECTED logic and resolve quickstart script syntax errors

* fix(ollama): align quickstart logic and cleanup test imports
This commit is contained in:
Md. Afzal Hassan Ehsani
2026-03-29 06:21:47 +05:30
committed by GitHub
parent c3c3075610
commit 905a4f3516
5 changed files with 328 additions and 13 deletions
Download Patch File Download Diff File Expand all files Collapse all files
core/framework/config.py
+7
View File
@@ -186,6 +186,8 @@ def get_worker_llm_extra_kwargs() -> dict[str, Any]:
"store": False,
"allowed_openai_params": ["store"],
}
if worker_llm.get("provider") == "ollama":
return {"num_ctx": worker_llm.get("num_ctx", 16384)}
return {}
@@ -432,6 +434,11 @@ def get_llm_extra_kwargs() -> dict[str, Any]:
"store": False,
"allowed_openai_params": ["store"],
}
if llm.get("provider") == "ollama":
# Pass num_ctx to Ollama so it doesn't silently truncate the ~9.5k Queen prompt.
# Ollama's default num_ctx is only 2048. We set it to 16384 here so LiteLLM
# passes it through as a provider-specific option.
return {"num_ctx": llm.get("num_ctx", 16384)}
return {}
core/framework/llm/litellm.py
+35 -1
View File
@@ -159,6 +159,26 @@ if litellm is not None:
# (e.g. stream_options for Anthropic) instead of forwarding them verbatim.
litellm.drop_params = True
def _is_ollama_model(model: str) -> bool:
"""Return True for any Ollama model string (ollama/ or ollama_chat/ prefix)."""
return model.startswith("ollama/") or model.startswith("ollama_chat/")
def _ensure_ollama_chat_prefix(model: str) -> str:
"""Normalise Ollama model strings to use the ollama_chat/ prefix.
LiteLLM requires the ``ollama_chat/`` prefix (not ``ollama/``) to enable
native function-calling support. With ``ollama/``, LiteLLM falls back to
JSON-mode tool calls, which the framework cannot parse as real tool calls.
See: https://docs.litellm.ai/docs/providers/ollama#example-usage---tool-calling
"""
if model.startswith("ollama/"):
return "ollama_chat/" + model[len("ollama/") :]
return model
RATE_LIMIT_MAX_RETRIES = 10
RATE_LIMIT_BACKOFF_BASE = 2 # seconds
RATE_LIMIT_MAX_DELAY = 120 # seconds - cap to prevent absurd waits
@@ -499,7 +519,9 @@ class LiteLLMProvider(LLMProvider):
# Translate kimi/ prefix to anthropic/ so litellm uses the Anthropic
# Messages API handler and routes to that endpoint — no special headers needed.
_original_model = model
if model.lower().startswith("kimi/"):
if _is_ollama_model(model):
model = _ensure_ollama_chat_prefix(model)
elif model.lower().startswith("kimi/"):
model = "anthropic/" + model[len("kimi/") :]
# Normalise api_base: litellm's Anthropic handler appends /v1/messages,
# so the base must be https://api.kimi.com/coding (no /v1 suffix).
@@ -722,6 +744,10 @@ class LiteLLMProvider(LLMProvider):
# Add tools if provided
if tools:
kwargs["tools"] = [self._tool_to_openai_format(t) for t in tools]
if _is_ollama_model(self.model):
# Ollama requires explicit tool_choice=auto for function calling
# so future readers don't have to guess.
kwargs.setdefault("tool_choice", "auto")
# Add response_format for structured output
# LiteLLM passes this through to the underlying provider
@@ -919,6 +945,10 @@ class LiteLLMProvider(LLMProvider):
kwargs["api_base"] = self.api_base
if tools:
kwargs["tools"] = [self._tool_to_openai_format(t) for t in tools]
if _is_ollama_model(self.model):
# Ollama requires explicit tool_choice=auto for function calling
# so future readers don't have to guess.
kwargs.setdefault("tool_choice", "auto")
if response_format:
kwargs["response_format"] = response_format
@@ -1620,6 +1650,10 @@ class LiteLLMProvider(LLMProvider):
kwargs["api_base"] = self.api_base
if tools:
kwargs["tools"] = [self._tool_to_openai_format(t) for t in tools]
if _is_ollama_model(self.model):
# Ollama requires explicit tool_choice=auto for function calling
# so future readers don't have to guess.
kwargs.setdefault("tool_choice", "auto")
if response_format:
kwargs["response_format"] = response_format
# The Codex ChatGPT backend (Responses API) rejects several params.
core/tests/test_litellm_provider.py
+105 -2
View File
@@ -18,11 +18,14 @@ from unittest.mock import AsyncMock, MagicMock, patch
import pytest
from framework.config import get_llm_extra_kwargs
from framework.llm.anthropic import AnthropicProvider
from framework.llm.litellm import (
OPENROUTER_TOOL_COMPAT_MODEL_CACHE,
LiteLLMProvider,
_compute_retry_delay,
_ensure_ollama_chat_prefix,
_is_ollama_model,
)
from framework.llm.provider import LLMProvider, LLMResponse, Tool
@@ -93,9 +96,9 @@ class TestLiteLLMProviderInit:
def test_init_ollama_no_key_needed(self):
"""Test that Ollama models don't require API key."""
with patch.dict(os.environ, {}, clear=True):
# Should not raise.
# Should not raise; ollama/ is normalised to ollama_chat/ for tool-call support.
provider = LiteLLMProvider(model="ollama/llama3")
assert provider.model == "ollama/llama3"
assert provider.model == "ollama_chat/llama3"
class TestLiteLLMProviderComplete:
@@ -1084,3 +1087,103 @@ class TestIsLocalModel:
from framework.runner.runner import AgentRunner
assert AgentRunner._is_local_model(model) is False
# ---------------------------------------------------------------------------
# Ollama helper functions
# ---------------------------------------------------------------------------
class TestIsOllamaModel:
"""Tests for _is_ollama_model()."""
@pytest.mark.parametrize(
"model",
[
"ollama/llama3",
"ollama/mistral:7b",
"ollama_chat/llama3",
"ollama_chat/qwen2.5:72b",
],
)
def test_ollama_models_return_true(self, model):
assert _is_ollama_model(model) is True
@pytest.mark.parametrize(
"model",
[
"gpt-4o-mini",
"anthropic/claude-3-haiku",
"openai/gpt-4o",
"gemini/gemini-1.5-flash",
"llama3",
"",
],
)
def test_non_ollama_models_return_false(self, model):
assert _is_ollama_model(model) is False
class TestEnsureOllamaChatPrefix:
"""Tests for _ensure_ollama_chat_prefix()."""
@pytest.mark.parametrize(
("input_model", "expected"),
[
("ollama/llama3", "ollama_chat/llama3"),
("ollama/mistral:7b", "ollama_chat/mistral:7b"),
("ollama/qwen2.5:72b-instruct", "ollama_chat/qwen2.5:72b-instruct"),
],
)
def test_rewrites_ollama_to_ollama_chat(self, input_model, expected):
assert _ensure_ollama_chat_prefix(input_model) == expected
@pytest.mark.parametrize(
"model",
[
"ollama_chat/llama3",
"gpt-4o-mini",
"anthropic/claude-3-haiku",
"gemini/gemini-1.5-flash",
"",
],
)
def test_leaves_non_ollama_prefix_unchanged(self, model):
assert _ensure_ollama_chat_prefix(model) == model
class TestGetLlmExtraKwargsOllama:
"""Tests for num_ctx injection via get_llm_extra_kwargs() for Ollama."""
def test_ollama_provider_returns_num_ctx(self):
"""Ollama config should inject num_ctx with default 16384."""
config = {
"llm": {"provider": "ollama", "model": "ollama/llama3"},
}
with patch("framework.config.get_hive_config", return_value=config):
result = get_llm_extra_kwargs()
assert result == {"num_ctx": 16384}
def test_ollama_provider_respects_custom_num_ctx(self):
"""User-specified num_ctx in config should take precedence."""
config = {
"llm": {"provider": "ollama", "model": "ollama/llama3", "num_ctx": 32768},
}
with patch("framework.config.get_hive_config", return_value=config):
result = get_llm_extra_kwargs()
assert result == {"num_ctx": 32768}
def test_non_ollama_provider_returns_empty(self):
"""Non-Ollama provider without subscriptions should return empty dict."""
config = {
"llm": {"provider": "anthropic", "model": "claude-3-haiku"},
}
with patch("framework.config.get_hive_config", return_value=config):
result = get_llm_extra_kwargs()
assert result == {}
def test_empty_config_returns_empty(self):
"""Missing config should return empty dict."""
with patch("framework.config.get_hive_config", return_value={}):
result = get_llm_extra_kwargs()
assert result == {}
quickstart.ps1
+96 -3
View File
@@ -1035,6 +1035,12 @@ $ProviderMenuUrls = @(
"https://openrouter.ai/keys"
)
$OllamaDetected = $false
try {
$null = & ollama list 2>$null
if ($LASTEXITCODE -eq 0) { $OllamaDetected = $true }
} catch { }
# ── Read previous configuration (if any) ──────────────────────
$PrevProvider = ""
$PrevModel = ""
@@ -1071,7 +1077,9 @@ if ($PrevSubMode -or $PrevProvider) {
"kimi_code" { if ($KimiCredDetected) { $prevCredValid = $true } }
"hive_llm" { if ($HiveCredDetected) { $prevCredValid = $true } }
default {
if ($PrevEnvVar) {
if ($PrevProvider -eq "ollama") {
$prevCredValid = $true
} elseif ($PrevEnvVar) {
$envVal = [System.Environment]::GetEnvironmentVariable($PrevEnvVar, "Process")
if (-not $envVal) { $envVal = [System.Environment]::GetEnvironmentVariable($PrevEnvVar, "User") }
if ($envVal) { $prevCredValid = $true }
@@ -1095,6 +1103,7 @@ if ($PrevSubMode -or $PrevProvider) {
"groq" { $DefaultChoice = "10" }
"cerebras" { $DefaultChoice = "11" }
"openrouter" { $DefaultChoice = "12" }
"ollama" { $DefaultChoice = "13" }
"minimax" { $DefaultChoice = "4" }
"kimi" { $DefaultChoice = "5" }
}
@@ -1163,7 +1172,17 @@ for ($idx = 0; $idx -lt $ProviderMenuEnvVars.Count; $idx++) {
if ($envVal) { Write-Color -Text " (credential detected)" -Color Green } else { Write-Host "" }
}
$SkipChoice = 7 + $ProviderMenuEnvVars.Count
# 13) Local (Ollama) — no API key needed
Write-Host " " -NoNewline
Write-Color -Text "13" -Color Cyan -NoNewline
if ($OllamaDetected) {
Write-Host ") Local (Ollama) - No API key needed " -NoNewline
Write-Color -Text "(ollama detected)" -Color Green
} else {
Write-Host ") Local (Ollama) - No API key needed"
}
$SkipChoice = 7 + $ProviderMenuEnvVars.Count + 1
Write-Host " " -NoNewline
Write-Color -Text "$SkipChoice" -Color Cyan -NoNewline
Write-Host ") Skip for now"
@@ -1383,6 +1402,75 @@ switch ($num) {
}
}
}
13 {
# Local (Ollama)
if (-not $OllamaDetected) {
Write-Host ""
Write-Warn "Ollama depends on a local Ollama server, but 'ollama list' failed."
Write-Host " Please install Ollama (https://ollama.com) and start the server,"
Write-Host " then run this quickstart again."
Write-Host ""
exit 1
}
$SelectedProviderId = "ollama"
Write-Host ""
Write-Ok "Using Local (Ollama)"
Write-Host ""
# Fetch available models
$ollamaModels = @()
try {
$listOutput = & ollama list 2>$null
if ($listOutput.Count -gt 1) {
for ($i = 1; $i -lt $listOutput.Count; $i++) {
$line = $listOutput[$i].Trim()
if ($line) {
$mName = ($line -split '\s+')[0]
if ($mName) { $ollamaModels += $mName }
}
}
}
} catch { }
if ($ollamaModels.Count -eq 0) {
Write-Warn "No Ollama models found."
Write-Host " Please open another terminal, run 'ollama run <model>' (e.g. 'ollama run llama3'),"
Write-Host " and then run this quickstart again."
Write-Host ""
exit 1
}
# Show model picker
Write-Host " Select an Ollama model:"
Write-Host ""
$defaultIdx = "1"
for ($i = 0; $i -lt $ollamaModels.Count; $i++) {
Write-Color -Text " $($i + 1)" -Color Cyan -NoNewline
Write-Host ") $($ollamaModels[$i])"
if ($PrevProvider -eq "ollama" -and $PrevModel -eq $ollamaModels[$i]) {
$defaultIdx = [string]($i + 1)
}
}
Write-Host ""
while ($true) {
$raw = Read-Host "Enter choice (1-$($ollamaModels.Count)) [$defaultIdx]"
if ([string]::IsNullOrWhiteSpace($raw)) { $raw = $defaultIdx }
if ($raw -match '^\d+$') {
$num = [int]$raw
if ($num -ge 1 -and $num -le $ollamaModels.Count) {
$SelectedModel = $ollamaModels[$num - 1]
Write-Host ""
Write-Ok "Model: $SelectedModel"
$SelectedMaxTokens = 8192
$SelectedMaxContextTokens = 16384
$SelectedApiBase = "http://localhost:11434"
break
}
}
Write-Color -Text "Invalid choice. Please enter 1-$($ollamaModels.Count)" -Color Red
}
}
{ $_ -eq $SkipChoice } {
Write-Host ""
Write-Warn "Skipped. An LLM API key is required to test and use worker agents."
@@ -1701,8 +1789,13 @@ if ($SelectedProviderId) {
} elseif ($SelectedProviderId -eq "openrouter") {
$config.llm["api_base"] = "https://openrouter.ai/api/v1"
$config.llm["api_key_env_var"] = $SelectedEnvVar
} else {
} elseif ($SelectedProviderId -eq "ollama") {
$config.llm["api_base"] = "http://localhost:11434"
$config.llm.Remove("api_key_env_var")
} elseif ($SelectedEnvVar) {
$config.llm["api_key_env_var"] = $SelectedEnvVar
} else {
$config.llm.Remove("api_key_env_var")
}
$config | ConvertTo-Json -Depth 4 | Set-Content -Path $HiveConfigFile -Encoding UTF8
quickstart.sh
+85 -7
View File
@@ -912,8 +912,9 @@ config["llm"] = {
"model": model,
"max_tokens": int(max_tokens),
"max_context_tokens": int(max_context_tokens),
"api_key_env_var": env_var,
}
if env_var:
config["llm"]["api_key_env_var"] = env_var
config["created_at"] = created_at
if use_claude_code_sub == "true":
@@ -1024,6 +1025,11 @@ elif [ -f "$HOME/.hive/antigravity-accounts.json" ]; then
ANTIGRAVITY_CRED_DETECTED=true
fi
OLLAMA_DETECTED=false
if ollama list >/dev/null 2>&1; then
OLLAMA_DETECTED=true
fi
# Detect API key providers
if [ "$USE_ASSOC_ARRAYS" = true ]; then
for env_var in "${!PROVIDER_NAMES[@]}"; do
@@ -1056,9 +1062,12 @@ try:
with open(cfg_path, encoding="utf-8-sig") as f:
c = json.load(f)
llm = c.get("llm", {})
print(f"PREV_PROVIDER={llm.get(\"provider\", \"\")}")
print(f"PREV_MODEL={llm.get(\"model\", \"\")}")
print(f"PREV_ENV_VAR={llm.get(\"api_key_env_var\", \"\")}")
prov = llm.get("provider", "")
mod = llm.get("model", "")
env = llm.get("api_key_env_var", "")
print(f"PREV_PROVIDER='{prov}'")
print(f"PREV_MODEL='{mod}'")
print(f"PREV_ENV_VAR='{env}'")
sub = ""
if llm.get("use_claude_code_subscription"):
sub = "claude_code"
@@ -1093,8 +1102,12 @@ if [ -n "$PREV_SUB_MODE" ] || [ -n "$PREV_PROVIDER" ]; then
hive_llm) [ "$HIVE_CRED_DETECTED" = true ] && PREV_CRED_VALID=true ;;
antigravity) [ "$ANTIGRAVITY_CRED_DETECTED" = true ] && PREV_CRED_VALID=true ;;
*)
# API key provider — check if the env var is set
if [ -n "$PREV_ENV_VAR" ] && [ -n "${!PREV_ENV_VAR}" ]; then
# API key provider — check if the env var is set; ollama uses local runtime detection
if [ "$PREV_PROVIDER" = "ollama" ]; then
if [ "$OLLAMA_DETECTED" = true ]; then
PREV_CRED_VALID=true
fi
elif [ -n "$PREV_ENV_VAR" ] && [ -n "${!PREV_ENV_VAR}" ]; then
PREV_CRED_VALID=true
fi
;;
@@ -1118,6 +1131,7 @@ if [ -n "$PREV_SUB_MODE" ] || [ -n "$PREV_PROVIDER" ]; then
groq) DEFAULT_CHOICE=11 ;;
cerebras) DEFAULT_CHOICE=12 ;;
openrouter) DEFAULT_CHOICE=13 ;;
ollama) DEFAULT_CHOICE=14 ;;
minimax) DEFAULT_CHOICE=4 ;;
kimi) DEFAULT_CHOICE=5 ;;
hive) DEFAULT_CHOICE=6 ;;
@@ -1196,7 +1210,14 @@ for idx in "${!PROVIDER_MENU_ENVS[@]}"; do
fi
done
SKIP_CHOICE=$((8 + ${#PROVIDER_MENU_ENVS[@]}))
# 14) Local (Ollama) — no API key needed
if [ "$OLLAMA_DETECTED" = true ]; then
echo -e " ${CYAN}14)${NC} Local (Ollama) - No API key needed ${GREEN}(ollama detected)${NC}"
else
echo -e " ${CYAN}14)${NC} Local (Ollama) - No API key needed"
fi
SKIP_CHOICE=$((8 + ${#PROVIDER_MENU_ENVS[@]} + 1))
echo -e " ${CYAN}$SKIP_CHOICE)${NC} Skip for now"
echo ""
@@ -1414,6 +1435,56 @@ case $choice in
PROVIDER_NAME="OpenRouter"
SIGNUP_URL="https://openrouter.ai/keys"
;;
14)
# Local (Ollama) — no API key; pick model from ollama list
if [ "$OLLAMA_DETECTED" != true ]; then
echo ""
echo -e "${YELLOW}Ollama depends on a local Ollama server, but 'ollama list' failed.${NC}"
echo -e " Please install Ollama (https://ollama.com) and start the server,"
echo -e " then run this quickstart again."
echo ""
exit 1
fi
SELECTED_PROVIDER_ID="ollama"
SELECTED_ENV_VAR=""
SELECTED_MAX_TOKENS=8192
SELECTED_MAX_CONTEXT_TOKENS=16384
OLLAMA_MODELS=()
while IFS= read -r line; do
[ -n "$line" ] && OLLAMA_MODELS+=("$line")
done < <(ollama list 2>/dev/null | tail -n +2 | awk '{print $1}')
if [ ${#OLLAMA_MODELS[@]} -gt 0 ]; then
echo ""
echo -e "${BOLD}Select an Ollama model:${NC}"
echo ""
for idx in "${!OLLAMA_MODELS[@]}"; do
num=$((idx + 1))
echo -e " ${CYAN}$num)${NC} ${OLLAMA_MODELS[$idx]}"
done
echo ""
while true; do
read -r -p "Enter choice (1-${#OLLAMA_MODELS[@]}): " model_choice
if [[ "$model_choice" =~ ^[0-9]+$ ]] && [ "$model_choice" -ge 1 ] && [ "$model_choice" -le ${#OLLAMA_MODELS[@]} ]; then
SELECTED_MODEL="${OLLAMA_MODELS[$((model_choice - 1))]}"
SELECTED_API_BASE="http://localhost:11434"
break
fi
echo -e "${RED}Invalid choice. Please enter 1-${#OLLAMA_MODELS[@]}${NC}"
done
echo ""
echo -e "${GREEN}⬢${NC} Using Ollama with model ${DIM}$SELECTED_MODEL${NC}"
echo -e "${YELLOW} ⚠ Note: The framework uses a ~9,500 token system prompt and requires strong tool use.${NC}"
echo -e "${YELLOW} For best results, use models like qwen2.5:72b+ or mistral-large.${NC}"
echo ""
else
echo ""
echo -e "${RED}No Ollama models found.${NC}"
echo -e " Please open another terminal, run ${CYAN}ollama pull llama3${NC} (or another model),"
echo -e " and then run this quickstart again."
echo ""
exit 1
fi
;;
"$SKIP_CHOICE")
echo ""
echo -e "${YELLOW}Skipped.${NC} An LLM API key is required to test and use worker agents."
@@ -1584,6 +1655,10 @@ if [ -n "$SELECTED_PROVIDER_ID" ]; then
save_configuration "$SELECTED_PROVIDER_ID" "$SELECTED_ENV_VAR" "$SELECTED_MODEL" "$SELECTED_MAX_TOKENS" "$SELECTED_MAX_CONTEXT_TOKENS" "" "$SELECTED_API_BASE" > /dev/null || SAVE_OK=false
elif [ "$SELECTED_PROVIDER_ID" = "openrouter" ]; then
save_configuration "$SELECTED_PROVIDER_ID" "$SELECTED_ENV_VAR" "$SELECTED_MODEL" "$SELECTED_MAX_TOKENS" "$SELECTED_MAX_CONTEXT_TOKENS" "" "$SELECTED_API_BASE" > /dev/null || SAVE_OK=false
elif [ "$SELECTED_PROVIDER_ID" = "ollama" ]; then
# Pass api_base explicitly — LiteLLM requires this to route ollama/* models
# to the local Ollama server instead of trying to reach a remote endpoint.
save_configuration "ollama" "" "$SELECTED_MODEL" "$SELECTED_MAX_TOKENS" "$SELECTED_MAX_CONTEXT_TOKENS" "" "http://localhost:11434" > /dev/null || SAVE_OK=false
else
save_configuration "$SELECTED_PROVIDER_ID" "$SELECTED_ENV_VAR" "$SELECTED_MODEL" "$SELECTED_MAX_TOKENS" "$SELECTED_MAX_CONTEXT_TOKENS" > /dev/null || SAVE_OK=false
fi
@@ -1859,6 +1934,9 @@ if [ -n "$SELECTED_PROVIDER_ID" ]; then
elif [ "$SELECTED_PROVIDER_ID" = "openrouter" ]; then
echo -e " ${GREEN}⬢${NC} OpenRouter API Key → ${DIM}$SELECTED_MODEL${NC}"
echo -e " ${DIM}API: openrouter.ai/api/v1 (OpenAI-compatible)${NC}"
elif [ "$SELECTED_PROVIDER_ID" = "ollama" ]; then
echo -e " ${GREEN}⬢${NC} Local (Ollama) → ${DIM}$SELECTED_MODEL${NC}"
echo -e " ${DIM}No API key required (runs locally via http://localhost:11434)${NC}"
else
echo -e " ${CYAN}$SELECTED_PROVIDER_ID${NC} → ${DIM}$SELECTED_MODEL${NC}"
fi