@oneciel-ai/claude-any 0.1.68 → 0.1.69

package/README.md

@@ -25,6 +25,12 @@
 
 ## Today's Top 3 Benefits
 
+### 2026-05-18
+
+1. **Realtime channel bridge for external agents** — Claude Any now exposes `/ca/channel/*` endpoints and an SSE connector so systems like AI-Net can push live agent messages into Claude Code sessions.
+2. **Advisor feedback is visible and actionable** — Advisor reviews can be surfaced in the Claude Code transcript and fed back into the executor model before plan approval or risky continuation points.
+3. **Better non-Anthropic workflow parity** — Cron-style task scheduling, channel polling, and router-native coordination commands are modeled more closely to Claude Code's native behavior for Ollama, Ollama Cloud, NVIDIA hosted, vLLM, and NIM.
+
 ### 2026-05-15
 
 1. **Router management is now navigable** — the built-in router home page is split into top-menu sections for Overview, LLM Settings, Events, and Endpoints instead of crowding everything into one long screen.
@@ -62,7 +68,7 @@ arguments through unchanged.
 
 Credits: One Ciel LLC
 
-Current version: `0.1.68`
+Current version: `0.1.69`
 
 ## Why This Exists
 
@@ -129,6 +135,33 @@ claude-any --ca-provider ollama-cloud --ca-model glm-5.1
 claude-any --ca-provider ollama --ca-base-url http://127.0.0.1:11434 --ca-model qwen3-coder
 ```
 
+Apply settings only, without launching Claude Code:
+
+```sh
+claude-any --ca-provider ollama-cloud --ca-model deepseek-v4-flash --ca-context-window 1048576 --ca-request-timeout-ms 300000 --ca-no-launch
+```
+
+Apply the recommended LLM options for the saved provider/model, then exit:
+
+```sh
+claude-any --ca-auto-llm-options --ca-no-launch
+```
+
+Apply the recommended LLM options for a specific model name, then exit:
+
+```sh
+claude-any --ca-provider ollama-cloud --ca-auto-llm-options deepseek-v4-flash --ca-no-launch
+```
+
+```sh
+claude-any --ca-env-file .env.claude-any --ca-no-launch
+```
+
+This is useful for remote provisioning, CI setup, base image preparation, or a
+parent agent that wants to configure Claude Any once and launch Claude Code in a
+separate step. Without `--ca-no-launch`, Claude Any applies the settings and
+then launches Claude Code as usual.
+
 Run one non-interactive Claude Code prompt:
 
 ```sh
@@ -147,6 +180,30 @@ Configure every launch option with flags:
 claude-any --ca-provider nvidia-hosted --ca-base-url https://integrate.api.nvidia.com/v1 --ca-model z-ai/glm-4.7 --ca-advisor-model deepseek-ai/deepseek-v4-pro --ca-api-key-env NVIDIA_API_KEY --ca-max-output-tokens 4096 --ca-context-window 65536 --ca-request-timeout-ms 120000 --ca-rate-limit-rpm 40 --ca-rate-limit-status on --ca-no-update-check -p "Reply with OK only." --output-format text
 ```
 
+Full Ollama Cloud 1M-context setup with flags:
+
+```sh
+claude-any --ca-provider ollama-cloud --ca-base-url https://ollama.com --ca-model deepseek-v4-flash --ca-advisor-model deepseek-v4-pro --ca-api-key-env OLLAMA_API_KEY --ca-context-window 1048576 --ca-max-output-tokens 8192 --ca-request-timeout-ms 300000 --ca-stream on --ca-stream-word-chunking off --ca-rate-limit-rpm 0 --ca-rate-limit-status on -p "Create an implementation plan." --output-format text
+```
+
+The same setup can let Claude Any choose the model's recommended LLM options:
+
+```sh
+claude-any --ca-provider ollama-cloud --ca-base-url https://ollama.com --ca-auto-llm-options deepseek-v4-flash --ca-advisor-model deepseek-v4-pro --ca-api-key-env OLLAMA_API_KEY --ca-stream on --ca-rate-limit-status on -p "Create an implementation plan." --output-format text
+```
+
+Full local Ollama setup with flags:
+
+```sh
+claude-any --ca-provider ollama --ca-base-url http://127.0.0.1:11434 --ca-model qwen3-coder --ca-ollama-num-ctx auto --ca-ollama-ctx-range 65536 262144 --ca-request-timeout-ms 180000 --ca-stream on --ca-no-update-check -p "Inspect the current project and summarize risks." --output-format text
+```
+
+Full NVIDIA hosted setup with rate-limit management:
+
+```sh
+claude-any --ca-provider nvidia-hosted --ca-base-url https://integrate.api.nvidia.com/v1 --ca-model z-ai/glm-4.7 --ca-advisor-model deepseek-ai/deepseek-v4-pro --ca-api-key-env NVIDIA_API_KEY --ca-context-window 65536 --ca-max-output-tokens 4096 --ca-request-timeout-ms 180000 --ca-stream on --ca-rate-limit-rpm 40 --ca-rate-limit-status on -p "Review this repository and list next actions." --output-format text
+```
+
 Or put the same values in environment variables:
 
 ```sh
@@ -164,18 +221,57 @@ export CLAUDE_ANY_RATE_LIMIT_STATUS=on
 claude-any -p "Reply with OK only." --output-format text
 ```
 
-For `.env` driven runs, save the same `CLAUDE_ANY_*` values in a file and pass
-it explicitly:
+For `.env` driven runs, save the same `CLAUDE_ANY_*` values in a file:
+
+```dotenv
+CLAUDE_ANY_SKIP_MENU=1
+CLAUDE_ANY_PROVIDER=ollama-cloud
+CLAUDE_ANY_BASE_URL=https://ollama.com
+CLAUDE_ANY_MODEL=deepseek-v4-flash
+CLAUDE_ANY_ADVISOR_MODEL=deepseek-v4-pro
+CLAUDE_ANY_API_KEY_ENV=OLLAMA_API_KEY
+CLAUDE_ANY_CONTEXT_WINDOW=1048576
+CLAUDE_ANY_MAX_OUTPUT_TOKENS=8192
+CLAUDE_ANY_REQUEST_TIMEOUT_MS=300000
+CLAUDE_ANY_STREAM=on
+CLAUDE_ANY_STREAM_WORD_CHUNKING=off
+CLAUDE_ANY_RATE_LIMIT_RPM=0
+CLAUDE_ANY_RATE_LIMIT_STATUS=on
+CLAUDE_ANY_WEB_SEARCH=on
+CLAUDE_ANY_WEB_FETCH=on
+CLAUDE_ANY_SELF_UPDATE_CHECK=off
+CLAUDE_ANY_UPDATE_CHECK=off
+```
+
+Then apply and launch:
 
 ```sh
 claude-any --ca-env-file .env.claude-any -p "Reply with OK only." --output-format text
 ```
 
+Or apply only and exit:
+
+```sh
+claude-any --ca-env-file .env.claude-any --ca-no-launch
+```
+
+You can also use a CLI flag to override one `.env` value for a single run:
+
+```sh
+claude-any --ca-env-file .env.claude-any --ca-model glm-5.1 -p "Use the overridden model for this run." --output-format text
+```
+
 Override order is deterministic: saved user choices from the menu are the
 baseline, OS environment variables override them, `--ca-env-file` values
 override the OS environment, CLI `--ca-*` parameters override the env file, and
 `--ca-menu` lets the final interactive menu choice override everything.
 
+Quietly upgrade Claude Any and Claude Code, then exit without launching Claude:
+
+```sh
+claude-any --ca-upgrade-and-exit
+```
+
 Headless coverage checklist: provider, base URL, model, Advisor model, API key
 or API-key environment variable, max output, context window, request timeout,
 RPM limit, RPM status display, streaming, web search, web fetch, Claude skills,
@@ -399,6 +495,17 @@ steps under that larger model's supervision.
 
 ## Changelog
 
+### 0.1.69
+
+- **Realtime channel bridge**: added `/ca/channel/messages`, `/ca/channel/wait`,
+  `/ca/channel/stream`, `/ca/channel/notify`, and runtime SSE connector controls
+  so external systems can deliver live messages into Claude Any.
+- **`/channel` slash command**: Claude Code can inspect bridge status, poll,
+  wait, send messages, and check SSE connector state without leaving the session.
+- **Advisor and coordination parity**: Advisor feedback is summarized back into
+  the visible transcript/executor flow, while channel and Cron-compatible tool
+  schemas make third-party model sessions behave closer to native Claude Code.
+
 ### 0.1.68
 
 - **Tabbed router management page**: the router root page now has top-menu

package/claude-any-tool-guard.py

@@ -740,9 +740,16 @@ def main() -> int:
     except Exception:
         return 0
     name = str(event.get("hook_event_name") or "")
+    provider = os.environ.get("CLAUDE_ANY_PROVIDER", "").strip()
+    is_active = active()
+    if not is_active:
+        if provider:
+            log_event(f"inactive provider={provider}")
+        return 0
 
-    # Worktree handlers always run, regardless of provider, so the non-git
-    # worktree fallback works whenever the hook is installed at all.
+    # Claude Any hooks are installed in Claude Code's global settings, so they
+    # must be silent for native Claude sessions. Only alter worktree, stop, and
+    # tool behavior when Claude Any launched the process with an active provider.
     if name == "WorktreeCreate":
         return handle_worktree_create(event)
     if name == "WorktreeRemove":
@@ -753,19 +760,13 @@ def main() -> int:
     # Lightweight observation for events we do not act on. Skip when inactive
     # to avoid touching disk on every event.
     if name in OBSERVE_ONLY_EVENTS:
-        if active():
-            try:
-                log_json_event(event)
-            except Exception:
-                pass
+        try:
+            log_json_event(event)
+        except Exception:
+            pass
         return 0
 
     # Tool/task events: keep existing provider gating.
-    provider = os.environ.get("CLAUDE_ANY_PROVIDER", "").strip()
-    if not active():
-        if provider:
-            log_event(f"inactive provider={provider}")
-        return 0
     if name == "PreToolUse":
         tool = str(event.get("tool_name") or "")
         raw = event.get("tool_input")