ayder-cli 1.7.3__tar.gz → 1.8.3__tar.gz

{ayder_cli-1.7.3 → ayder_cli-1.8.3}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: ayder-cli
-Version: 1.7.3
+Version: 1.8.3
 Summary: AI agent for any LLMs
 Project-URL: Homepage, https://github.com/ayder/ayder-cli
 Project-URL: Repository, https://github.com/ayder/ayder-cli.git
@@ -19,7 +19,7 @@ Requires-Dist: google-genai
 Requires-Dist: httpx
 Requires-Dist: libcst>=1.0.0
 Requires-Dist: loguru
-Requires-Dist: ollama>=0.6.1
+Requires-Dist: ollama>=0.6.2
 Requires-Dist: openai
 Requires-Dist: python-dotenv>=1.0.0
 Requires-Dist: rich>=13.0.0
@@ -54,39 +54,11 @@ Most AI coding assistants require cloud APIs, subscriptions, or heavy IDE plugin
 - **Agentic workflow** -- the LLM reads files, edits code, runs shell commands, and iterates autonomously with configurable iteration limits per message.
 - **Multi-agent** -- define specialized sub-agents in `config.toml`. Each agent runs independently with its own LLM, model, and context. Results are injected back into the main conversation when complete.
 - **Textual TUI** -- an inline terminal interface with chat view, tool panel, thinking block toggle, slash command auto-completion, permission toggles, and tool confirmation modals with diff previews.
-- **Minimal dependencies** -- OpenAI SDK, Rich, and Textual. Other provider SDKs are optional.
-
-### Tested Providers with Models
-
-| Provider | Location | Model |
-| -------- | -------- | ----- |
-| ollama | Cloud | deepseek-v3.2:cloud |
-| ollama | Cloud | gemini-3-pro-preview:latest |
-| ollama | Local | glm-4.7-flash:latest |
-| ollama | Cloud | glm-4.7:cloud |
-| ollama | Cloud | glm-5:cloud |
-| ollama | Local | glm-ocr:latest |
-| ollama | Cloud | gpt-oss:120b-cloud |
-| ollama | Cloud | kimi-k2.5:cloud |
-| ollama | Cloud | minimax-m2.5:cloud |
-| ollama | Local | ministral-3:14b |
-| ollama | Cloud | qwen3-coder-next:cloud |
-| ollama | Cloud | qwen3-coder:480b-cloud |
-| ollama | Local | qwen3-coder:latest |
-| anthropic | Cloud | claude-opus-4-6 |
-| anthropic | Cloud | claude-sonnet-4-5-20250929 |
-| anthropic | Cloud | claude-haiku-4-5-20251001 |
-| openai | Cloud | GPT-5.3-Codex |
-| openai | Cloud | GPT-5.3-Codex-Spark |
-| openai | Cloud | GPT-5.2 |
-| openai | Cloud | GPT-5 |
-| gemini | Cloud | gemini-3-deep-think |
-| gemini | Cloud | gemini-3-pro |
-| gemini | Cloud | gemini-3-flash |
+- **Batteries included** -- the OpenAI, Anthropic, Google, and Ollama SDKs all ship with ayder-cli, so every provider works out of the box. The terminal UI is built on Rich and Textual.
 
 ### Tools
 
-LLMs on their own can only generate text. To be a useful coding assistant, the model needs to *act* on your codebase. ayder-cli provides 25 tools across 10 categories that the model can call:
+LLMs on their own can only generate text. To be a useful coding assistant, the model needs to *act* on your codebase. ayder-cli ships 16 built-in tools that the model can call, plus optional plugin toolsets (see [Plugins](#plugins)):
 
 Each tool has an OpenAI-compatible JSON schema so models that support function calling can use them natively. For models that don't, ayder-cli also parses a custom XML-like syntax (`<function=name><parameter=key>value</parameter></function>`) as a fallback.
 
@@ -134,26 +106,16 @@ export OLLAMA_FLASH_ATTENTION=true
 export OLLAMA_MAX_LOADED_MODELS=1
 ```
 
-### Anthropic setup (optional)
+### Cloud providers (Anthropic, OpenAI, Gemini)
 
-```bash
-# Install the Anthropic SDK
-pip install anthropic
-
-# Set your API key in ~/.ayder/config.toml (see Configuration below)
-# Then switch provider:
-#   /provider anthropic
-```
-
-### Gemini setup (optional)
+The Anthropic, OpenAI, and Google SDKs ship with ayder-cli — no extra installs.
+Just add an API key to a profile in `~/.ayder/config.toml` (see Configuration
+below) and switch provider in the TUI:
 
 ```bash
-# Install the Google Generative AI SDK
-pip install google-generativeai
+/provider anthropic   # or: openai, gemini
 ```
 
-Set your API key in `~/.ayder/config.toml`, then switch provider: `/provider gemini`
-
 ### Configuration: Profiles and Drivers
 
 ayder-cli uses a flexible profile-based configuration system. On the first run, it creates a config file at `~/.ayder/config.toml`.
@@ -218,7 +180,7 @@ rotation = "10 MB"
 retention = "7 days"
 
 [context_manager]
-enabled = false
+enabled = true
 max_context_tokens = 8192
 
 [temporal]
@@ -316,12 +278,14 @@ ayder --implement-all
 
 By default, every tool call requires user confirmation. Use permission flags to auto-approve tool categories:
 
-| Flag | Category | Tools |
-| ---- | -------- | ----- |
-| `-r` | Read | `file_explorer`, `read_file`, `search_codebase`, `get_project_structure`, `load_memory`, `get_background_output`, `list_background_processes`, `list_tasks`, `show_task`, `list_virtualenvs`, `activate_virtualenv` |
-| `-w` | Write | `file_editor`, `create_note`, `save_memory`, `manage_environment_vars`, `python_editor`, `temporal_workflow` |
-| `-x` | Execute | `run_shell_command`, `run_background_process`, `kill_background_process`, `create_virtualenv`, `install_requirements`, `remove_virtualenv` |
-| `--http` | Web/Network | `fetch_web`, `dbs_tool` |
+| Flag | Category | Built-in tools |
+| ---- | -------- | -------------- |
+| `-r` | Read | `file_explorer`, `read_file`, `search_codebase`, `get_project_structure`, `get_background_output`, `list_background_processes`, `list_tasks`, `show_task` |
+| `-w` | Write | `file_editor`, `context`, `create_note`, `manage_environment_vars` |
+| `-x` | Execute | `run_shell_command`, `run_background_process`, `kill_background_process` |
+| `--http` | Web/Network | `fetch_web` |
+
+Installed plugins extend these categories (e.g. `python_editor` → write, virtualenv tools → execute, `dbs_tool` → web).
 
 ```bash
 # Auto-approve read-only tools
@@ -370,7 +334,7 @@ The context manager is configured under `[context_manager]` in `config.toml`:
 | `max_tool_result_length` | `2048` | Maximum character length for a compressed tool result. |
 | `compress_tool_results` | `true` | Enable automatic tool result compression. |
 
-You can also manually manage context with `/save-memory`, `/load-memory`, and `/compact`.
+You can also manually manage context with the `context` tool and its slash-command shortcuts: `/save-context`, `/load-context`, `/list-contexts`, `/context-stats`, and `/compact`.
 
 ### Slash Commands
 
@@ -390,9 +354,11 @@ You can also manually manage context with `/save-memory`, `/load-memory`, and `/
 | `/skill` | Activate a domain skill from `.ayder/skills/` |
 | `/verbose` | Toggle verbose mode |
 | `/logging` | Set log level for current session (NONE, ERROR, WARNING, INFO, DEBUG) |
-| `/compact` | Summarize conversation, save to memory, clear, and reload context |
-| `/save-memory` | Summarize conversation and save to memory (no clear) |
-| `/load-memory` | Load memory and restore context |
+| `/compact` | Summarize conversation, save a context snapshot, clear, and reload |
+| `/save-context` | Summarize conversation and save it to a named context slot (no clear) |
+| `/load-context` | Load a saved context slot and restore it |
+| `/list-contexts` | List saved context slots |
+| `/context-stats` | Show current token and KV-cache usage |
 | `/archive-completed-tasks` | Move completed tasks to `.ayder/task_archive/` |
 | `/temporal` | Start/status Temporal queue worker |
 | `/agent list` | List configured agents and their current status |
@@ -436,7 +402,7 @@ The standard mode for general coding and chat. Uses the system prompt.
 > create a fibonacci function
 ```
 
-Available tools: File read/write, shell commands, search, memory, notes, tasks.
+Available tools: file read/write, shell commands, search, context, notes, tasks.
 
 ### Planning Mode (`/plan`)
 
@@ -486,24 +452,23 @@ The tool system:
 - Supports tag-based filtering for dynamic enable/disable
 - Injects tool-specific system prompts when enabled
 
-**Current tool categories (25 tools):**
+**Built-in tools (16):**
 
 | Category | Tools |
 |----------|-------|
 | **Filesystem** | `file_explorer`, `read_file`, `file_editor` |
 | **Search** | `search_codebase`, `get_project_structure` |
 | **Shell** | `run_shell_command` |
-| **Python Editor** | `python_editor` (CST-based structural code manipulation) |
-| **Memory** | `save_memory`, `load_memory` |
+| **Context** | `context` (save / load / list / stats / clear session context slots) |
 | **Notes** | `create_note` |
 | **Background Processes** | `run_background_process`, `get_background_output`, `kill_background_process`, `list_background_processes` |
 | **Tasks** | `list_tasks`, `show_task` |
 | **Environment** | `manage_environment_vars` |
-| **Virtual Environments** | `create_virtualenv`, `install_requirements`, `list_virtualenvs`, `activate_virtualenv`, `remove_virtualenv` |
 | **Web** | `fetch_web` |
-| **DBS** | `dbs_tool` (RAG API for DBS-related queries) |
-| **Workflow** | `temporal_workflow` |
-| **Agents** | `call_agent` (dispatch a named agent to run a task in the background) |
+
+When agents are configured, the `call_agent` tool is also registered so the main LLM can dispatch them (see [Multi-Agent System](#multi-agent-system)).
+
+**Plugin tools** — `python_editor`, `dbs_tool`, `temporal_workflow`, virtualenv tools, and MCP integrations ship as optional plugins in the [ayder/ayder-plugins](https://github.com/ayder/ayder-plugins) repo (see [Plugins](#plugins)). Once installed they are auto-discovered exactly like built-ins.
 
 ## Multi-Agent System
 

{ayder_cli-1.7.3 → ayder_cli-1.8.3}/README.md

@@ -25,39 +25,11 @@ Most AI coding assistants require cloud APIs, subscriptions, or heavy IDE plugin
 - **Agentic workflow** -- the LLM reads files, edits code, runs shell commands, and iterates autonomously with configurable iteration limits per message.
 - **Multi-agent** -- define specialized sub-agents in `config.toml`. Each agent runs independently with its own LLM, model, and context. Results are injected back into the main conversation when complete.
 - **Textual TUI** -- an inline terminal interface with chat view, tool panel, thinking block toggle, slash command auto-completion, permission toggles, and tool confirmation modals with diff previews.
-- **Minimal dependencies** -- OpenAI SDK, Rich, and Textual. Other provider SDKs are optional.
-
-### Tested Providers with Models
-
-| Provider | Location | Model |
-| -------- | -------- | ----- |
-| ollama | Cloud | deepseek-v3.2:cloud |
-| ollama | Cloud | gemini-3-pro-preview:latest |
-| ollama | Local | glm-4.7-flash:latest |
-| ollama | Cloud | glm-4.7:cloud |
-| ollama | Cloud | glm-5:cloud |
-| ollama | Local | glm-ocr:latest |
-| ollama | Cloud | gpt-oss:120b-cloud |
-| ollama | Cloud | kimi-k2.5:cloud |
-| ollama | Cloud | minimax-m2.5:cloud |
-| ollama | Local | ministral-3:14b |
-| ollama | Cloud | qwen3-coder-next:cloud |
-| ollama | Cloud | qwen3-coder:480b-cloud |
-| ollama | Local | qwen3-coder:latest |
-| anthropic | Cloud | claude-opus-4-6 |
-| anthropic | Cloud | claude-sonnet-4-5-20250929 |
-| anthropic | Cloud | claude-haiku-4-5-20251001 |
-| openai | Cloud | GPT-5.3-Codex |
-| openai | Cloud | GPT-5.3-Codex-Spark |
-| openai | Cloud | GPT-5.2 |
-| openai | Cloud | GPT-5 |
-| gemini | Cloud | gemini-3-deep-think |
-| gemini | Cloud | gemini-3-pro |
-| gemini | Cloud | gemini-3-flash |
+- **Batteries included** -- the OpenAI, Anthropic, Google, and Ollama SDKs all ship with ayder-cli, so every provider works out of the box. The terminal UI is built on Rich and Textual.
 
 ### Tools
 
-LLMs on their own can only generate text. To be a useful coding assistant, the model needs to *act* on your codebase. ayder-cli provides 25 tools across 10 categories that the model can call:
+LLMs on their own can only generate text. To be a useful coding assistant, the model needs to *act* on your codebase. ayder-cli ships 16 built-in tools that the model can call, plus optional plugin toolsets (see [Plugins](#plugins)):
 
 Each tool has an OpenAI-compatible JSON schema so models that support function calling can use them natively. For models that don't, ayder-cli also parses a custom XML-like syntax (`<function=name><parameter=key>value</parameter></function>`) as a fallback.
 
@@ -105,26 +77,16 @@ export OLLAMA_FLASH_ATTENTION=true
 export OLLAMA_MAX_LOADED_MODELS=1
 ```
 
-### Anthropic setup (optional)
+### Cloud providers (Anthropic, OpenAI, Gemini)
 
-```bash
-# Install the Anthropic SDK
-pip install anthropic
-
-# Set your API key in ~/.ayder/config.toml (see Configuration below)
-# Then switch provider:
-#   /provider anthropic
-```
-
-### Gemini setup (optional)
+The Anthropic, OpenAI, and Google SDKs ship with ayder-cli — no extra installs.
+Just add an API key to a profile in `~/.ayder/config.toml` (see Configuration
+below) and switch provider in the TUI:
 
 ```bash
-# Install the Google Generative AI SDK
-pip install google-generativeai
+/provider anthropic   # or: openai, gemini
 ```
 
-Set your API key in `~/.ayder/config.toml`, then switch provider: `/provider gemini`
-
 ### Configuration: Profiles and Drivers
 
 ayder-cli uses a flexible profile-based configuration system. On the first run, it creates a config file at `~/.ayder/config.toml`.
@@ -189,7 +151,7 @@ rotation = "10 MB"
 retention = "7 days"
 
 [context_manager]
-enabled = false
+enabled = true
 max_context_tokens = 8192
 
 [temporal]
@@ -287,12 +249,14 @@ ayder --implement-all
 
 By default, every tool call requires user confirmation. Use permission flags to auto-approve tool categories:
 
-| Flag | Category | Tools |
-| ---- | -------- | ----- |
-| `-r` | Read | `file_explorer`, `read_file`, `search_codebase`, `get_project_structure`, `load_memory`, `get_background_output`, `list_background_processes`, `list_tasks`, `show_task`, `list_virtualenvs`, `activate_virtualenv` |
-| `-w` | Write | `file_editor`, `create_note`, `save_memory`, `manage_environment_vars`, `python_editor`, `temporal_workflow` |
-| `-x` | Execute | `run_shell_command`, `run_background_process`, `kill_background_process`, `create_virtualenv`, `install_requirements`, `remove_virtualenv` |
-| `--http` | Web/Network | `fetch_web`, `dbs_tool` |
+| Flag | Category | Built-in tools |
+| ---- | -------- | -------------- |
+| `-r` | Read | `file_explorer`, `read_file`, `search_codebase`, `get_project_structure`, `get_background_output`, `list_background_processes`, `list_tasks`, `show_task` |
+| `-w` | Write | `file_editor`, `context`, `create_note`, `manage_environment_vars` |
+| `-x` | Execute | `run_shell_command`, `run_background_process`, `kill_background_process` |
+| `--http` | Web/Network | `fetch_web` |
+
+Installed plugins extend these categories (e.g. `python_editor` → write, virtualenv tools → execute, `dbs_tool` → web).
 
 ```bash
 # Auto-approve read-only tools
@@ -341,7 +305,7 @@ The context manager is configured under `[context_manager]` in `config.toml`:
 | `max_tool_result_length` | `2048` | Maximum character length for a compressed tool result. |
 | `compress_tool_results` | `true` | Enable automatic tool result compression. |
 
-You can also manually manage context with `/save-memory`, `/load-memory`, and `/compact`.
+You can also manually manage context with the `context` tool and its slash-command shortcuts: `/save-context`, `/load-context`, `/list-contexts`, `/context-stats`, and `/compact`.
 
 ### Slash Commands
 
@@ -361,9 +325,11 @@ You can also manually manage context with `/save-memory`, `/load-memory`, and `/
 | `/skill` | Activate a domain skill from `.ayder/skills/` |
 | `/verbose` | Toggle verbose mode |
 | `/logging` | Set log level for current session (NONE, ERROR, WARNING, INFO, DEBUG) |
-| `/compact` | Summarize conversation, save to memory, clear, and reload context |
-| `/save-memory` | Summarize conversation and save to memory (no clear) |
-| `/load-memory` | Load memory and restore context |
+| `/compact` | Summarize conversation, save a context snapshot, clear, and reload |
+| `/save-context` | Summarize conversation and save it to a named context slot (no clear) |
+| `/load-context` | Load a saved context slot and restore it |
+| `/list-contexts` | List saved context slots |
+| `/context-stats` | Show current token and KV-cache usage |
 | `/archive-completed-tasks` | Move completed tasks to `.ayder/task_archive/` |
 | `/temporal` | Start/status Temporal queue worker |
 | `/agent list` | List configured agents and their current status |
@@ -407,7 +373,7 @@ The standard mode for general coding and chat. Uses the system prompt.
 > create a fibonacci function
 ```
 
-Available tools: File read/write, shell commands, search, memory, notes, tasks.
+Available tools: file read/write, shell commands, search, context, notes, tasks.
 
 ### Planning Mode (`/plan`)
 
@@ -457,24 +423,23 @@ The tool system:
 - Supports tag-based filtering for dynamic enable/disable
 - Injects tool-specific system prompts when enabled
 
-**Current tool categories (25 tools):**
+**Built-in tools (16):**
 
 | Category | Tools |
 |----------|-------|
 | **Filesystem** | `file_explorer`, `read_file`, `file_editor` |
 | **Search** | `search_codebase`, `get_project_structure` |
 | **Shell** | `run_shell_command` |
-| **Python Editor** | `python_editor` (CST-based structural code manipulation) |
-| **Memory** | `save_memory`, `load_memory` |
+| **Context** | `context` (save / load / list / stats / clear session context slots) |
 | **Notes** | `create_note` |
 | **Background Processes** | `run_background_process`, `get_background_output`, `kill_background_process`, `list_background_processes` |
 | **Tasks** | `list_tasks`, `show_task` |
 | **Environment** | `manage_environment_vars` |
-| **Virtual Environments** | `create_virtualenv`, `install_requirements`, `list_virtualenvs`, `activate_virtualenv`, `remove_virtualenv` |
 | **Web** | `fetch_web` |
-| **DBS** | `dbs_tool` (RAG API for DBS-related queries) |
-| **Workflow** | `temporal_workflow` |
-| **Agents** | `call_agent` (dispatch a named agent to run a task in the background) |
+
+When agents are configured, the `call_agent` tool is also registered so the main LLM can dispatch them (see [Multi-Agent System](#multi-agent-system)).
+
+**Plugin tools** — `python_editor`, `dbs_tool`, `temporal_workflow`, virtualenv tools, and MCP integrations ship as optional plugins in the [ayder/ayder-plugins](https://github.com/ayder/ayder-plugins) repo (see [Plugins](#plugins)). Once installed they are auto-discovered exactly like built-ins.
 
 ## Multi-Agent System
 

{ayder_cli-1.7.3 → ayder_cli-1.8.3}/docs/PROJECT_STRUCTURE.md

@@ -289,7 +289,7 @@ the design rationale.
 | `tools/builtins/filesystem.py` | File system tool impls | `file_explorer()`, `read_file()`, `file_editor()` |
 | `tools/builtins/search.py` | Search tool impls | `search_codebase()`, `get_project_structure()` |
 | `tools/builtins/shell.py` | Shell execution impl | `run_shell_command()` |
-| `tools/builtins/memory.py` | Long-term memory tools | `save_memory`, `load_memory`, `save_context_memory`, `load_context_memory` |
+| `tools/builtins/context.py` | Session context snapshots | `context` |
 | `tools/builtins/notes.py` | Note-taking tool | `create_note()` |
 | `tools/builtins/tasks.py` | Task-management tools | `list_tasks()`, `show_task()` |
 | `tools/builtins/web.py` | HTTP fetch tool | `fetch_web()` |
@@ -642,11 +642,11 @@ Schema-driven tool definitions with auto-discovery:
 - **Schema generation**: `to_openai_schema()` returns the OpenAI function-calling dict
 - **Plugin loading**: `tools/plugin_manager.py` can augment `TOOL_DEFINITIONS` at runtime from local or GitHub-sourced plugins
 
-19 built-in tools across 9 definition files:
+16 built-in tools across 9 definition files:
 - Filesystem (3): `file_explorer`, `read_file`, `file_editor`
 - Search (2): `search_codebase`, `get_project_structure`
 - Shell (1): `run_shell_command`
-- Memory (4): `save_memory`, `load_memory`, `save_context_memory`, `load_context_memory`
+- Context (1): `context`
 - Notes (1): `create_note`
 - Background Processes (4): `run_background_process`, `get_background_output`, `kill_background_process`, `list_background_processes`
 - Tasks (2): `list_tasks`, `show_task`

{ayder_cli-1.7.3 → ayder_cli-1.8.3}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "ayder-cli"
-version = "1.7.3"
+version = "1.8.3"
 description = "AI agent for any LLMs"
 readme = "README.md"
 requires-python = ">=3.12"
@@ -23,7 +23,7 @@ dependencies = [
     "openai",
     "anthropic",
     "google-genai",
-    "ollama>=0.6.1",
+    "ollama>=0.6.2",
     "loguru",
     "httpx",
     "rich>=13.0.0",

{ayder_cli-1.7.3 → ayder_cli-1.8.3}/src/ayder_cli/application/runtime_factory.py

@@ -62,7 +62,14 @@ def create_runtime(
     llm_provider: AIProvider = provider_orchestrator.create(cfg)
     project_ctx = ProjectContext(project_root)
     process_manager = ProcessManager(max_processes=cfg.max_background_processes)
-    tool_registry = create_default_registry(project_ctx, process_manager=process_manager)
+
+    # Create context manager before registry so tools can receive it via DI.
+    context_mgr = context_manager_factory.create(cfg)
+    tool_registry = create_default_registry(
+        project_ctx,
+        process_manager=process_manager,
+        context_manager=context_mgr,
+    )
 
     try:
         structure = tool_registry.execute("get_project_structure", {"max_depth": 3})
@@ -76,8 +83,7 @@ def create_runtime(
     tool_prompts = tool_registry.get_system_prompts(tags=tool_tags)
     system_prompt = base_prompt + tool_prompts + macro
 
-    # Create context manager and freeze system prompt
-    context_mgr = context_manager_factory.create(cfg)
+    # Freeze system prompt after tool schemas are known.
     tool_schemas = tool_registry.get_schemas(tags=tool_tags)
     context_mgr.freeze_system_prompt(system_prompt, tool_schemas)
 
@@ -161,8 +167,13 @@ def create_agent_runtime(
     # 3. Create isolated AIProvider
     llm_provider = provider_orchestrator.create(cfg)
 
-    # 4. Create isolated ToolRegistry (shared PM and ProjectContext)
-    tool_registry = create_default_registry(project_ctx, process_manager=process_manager)
+    # 4. Create isolated context manager and ToolRegistry (shared PM/ProjectContext)
+    context_mgr = context_manager_factory.create(cfg)
+    tool_registry = create_default_registry(
+        project_ctx,
+        process_manager=process_manager,
+        context_manager=context_mgr,
+    )
 
     # 5. Build agent system prompt
     identity_prefix = (
@@ -182,8 +193,7 @@ def create_agent_runtime(
     tool_prompts = tool_registry.get_system_prompts(tags=tool_tags)
     system_prompt = identity_prefix + agent_config.system_prompt + tool_prompts + summary_suffix
 
-    # Create context manager and freeze agent system prompt
-    context_mgr = context_manager_factory.create(cfg)
+    # Freeze agent system prompt after tool schemas are known.
     tool_schemas = tool_registry.get_schemas(tags=tool_tags)
     context_mgr.freeze_system_prompt(system_prompt, tool_schemas)
 

{ayder_cli-1.7.3 → ayder_cli-1.8.3}/src/ayder_cli/parser.py

@@ -95,6 +95,13 @@ class ContentProcessor:
     _RE_ORPHAN_PARAMETER = re.compile(
         r"</?\uff5c?\uff24?\uff33?\uff2d?\uff2c?\uff5c?parameter[^>]*>"
     )
+    # MiniMax models use special structural tokens that can leak into
+    # msg.content when an IN_CONTENT driver is used on a multi-turn request
+    # (observed empirically on minimax-m3:cloud as ]<]minimax[>[ ). Older M2
+    # builds documented the ]~b]<role> / [e~[ role markers. Strip all variants.
+    _RE_ORPHAN_MINIMAX_MARKER = re.compile(
+        r"\]<\]minimax\[>\[|\]~b\](?:system|user|ai|tool)?|\[e~\["
+    )
 
     # -- JSON tool arrays (for stripping and parsing) ------------------------
     _RE_JSON_TOOL_ARRAY = re.compile(r'\[\s*\{[^}]*"function"\s*:.*?\}\s*\]', re.DOTALL)
@@ -427,6 +434,7 @@ class ContentProcessor:
         text = self._RE_ORPHAN_FUNCTION_CALLS.sub("", text)
         text = self._RE_ORPHAN_INVOKE.sub("", text)
         text = self._RE_ORPHAN_PARAMETER.sub("", text)
+        text = self._RE_ORPHAN_MINIMAX_MARKER.sub("", text)
         return text
 
     def _strip_json_tool_arrays(self, text: str) -> str:

{ayder_cli-1.7.3 → ayder_cli-1.8.3}/src/ayder_cli/prompts.py

@@ -174,44 +174,9 @@ If you encounter errors, fix them and continue. Only stop when all tasks are mar
 
 
 # =============================================================================
-# MEMORY & CONVERSATION MANAGEMENT PROMPTS (commands/system.py)
+# CONVERSATION MANAGEMENT PROMPTS
 # =============================================================================
-# Used by: commands/system.py::ClearCommand.execute()
-# REASON: After clearing conversation history, confirm the LLM understands
-# the context is fresh and ready for new work.
-
-CLEAR_COMMAND_RESET_PROMPT = """The conversation has been cleared. Please acknowledge this reset and confirm you're ready to start fresh."""
-
-
-# Used by: commands/system.py::SummaryCommand.execute()
-# REASON: Extract key decisions and context from a long conversation and
-# persist it to a memory file for future reference or session restoration.
-
-SUMMARY_COMMAND_PROMPT_TEMPLATE = """Please summarize the following conversation and save it to `.ayder/current_memory.md`.
-
-The summary should:
-- Capture key decisions, context, and progress
-- Be concise but comprehensive
-- Help future me understand what we've done so far
-
-Conversation:
-{conversation_text}
-
-Use the write_file tool to save the summary to `.ayder/current_memory.md`."""
-
-
-# Used by: commands/system.py::LoadCommand.execute()
-# REASON: Restore context from a previously saved memory file so the LLM
-# can continue work from where it left off in a previous session.
-
-LOAD_MEMORY_COMMAND_PROMPT_TEMPLATE = """I've loaded the previous conversation memory from `.ayder/current_memory.md`. 
-Please acknowledge this context and continue from where we left off.
-
-[Memory from previous conversation]:
-{memory_content}"""
-
-
-# Used by: commands/system.py::CompactCommand.execute()
+# Used by: tui/commands.py::handle_compact()
 # REASON: Combine summary, save, clear, and reload into a single operation
 # to prevent context window overflow while maintaining continuity.
 
@@ -227,24 +192,21 @@ Provide a brief summary, save it, confirm reset, and acknowledge the context."""
 
 
 # =============================================================================
-# SAVE/LOAD MEMORY COMMANDS (tui/commands.py)
+# SAVE CONTEXT COMMAND (tui/commands.py)
 # =============================================================================
-# Used by: tui/commands.py::handle_save_memory()
-# REASON: Prompt LLM to create and save a memory summary without clearing
-# the conversation history.
-
-SAVE_MEMORY_COMMAND_PROMPT_TEMPLATE = """Please create a memory summary of our conversation:
-
-{conversation_text}
+# Used by: tui/commands.py::handle_save_context()
 
-Save this summary to `.ayder/memory/current_memory.md` using the write_file tool.
+SAVE_CONTEXT_COMMAND_PROMPT_TEMPLATE = """Summarize the following conversation in 200-400 words, focusing on:
+- Goals and decisions made
+- Code changes attempted or completed
+- Open questions or next steps
 
-The summary should:
-- Capture key decisions, context, and progress
-- Be concise but comprehensive
-- Help future sessions understand what we've done so far
+Then call the context tool exactly once to save it:
+  context(action="save", name="{name}", content="<your summary>")
 
-Save the summary and confirm when done."""
+Conversation:
+{conversation_text}
+"""
 
 
 # =============================================================================
@@ -267,15 +229,3 @@ the skill content to determine when to apply it.
 ### END SKILL
 """
 
-
-# Used by: tui/commands.py::handle_load_memory()
-# REASON: Prompt LLM to load and acknowledge memory from file.
-
-LOAD_MEMORY_COMMAND_PROMPT_TEMPLATE = """Please load the previous memory from `.ayder/memory/current_memory.md` and acknowledge the context.
-
-Previous context:
-```
-{memory_content}
-```
-
-Please acknowledge this context and continue from where we left off."""

{ayder_cli-1.7.3 → ayder_cli-1.8.3}/src/ayder_cli/providers/impl/ollama_drivers/matrix.py

@@ -76,10 +76,30 @@ RESOLUTION_MATRIX: tuple[ResolutionRule, ...] = (
         driver="generic_native",
         note="Native works on current Ollama; verified for deepseek-v4-pro:cloud",
     ),
+    # minimax: native tool extraction verified working on current Ollama
+    # (probed minimax-m3:cloud — returns clean msg.tool_calls, empty content,
+    # both single- and multi-turn). The IN_CONTENT MiniMaxDriver leaks the
+    # model's native tool-call special token (]<]minimax[>[) on multi-turn
+    # requests: _convert_messages feeds prior tool_calls/results back, Ollama's
+    # template renders them with native tokens, the model mimics them, and the
+    # text parser neither parses nor strips them — garbling the call. Trusting
+    # native; reactive fallback engages on #14834. If a specific minimax variant
+    # regresses, prefer a model-name-specific driver (priority < 50) over
+    # re-engaging the family-wide IN_CONTENT path.
     ResolutionRule(
         family_substring="minimax",
-        driver="minimax",
-        note="Emits namespaced minimax tool_call blocks in message content",
+        driver="generic_native",
+        note="Native works on current Ollama; verified for minimax-m3:cloud",
+    ),
+    # kimi: native tool extraction verified working on current Ollama (probed
+    # kimi-k2.6:cloud — clean msg.tool_calls, empty content, single- and
+    # multi-turn). NOT affected by ollama#14360 (Kimi K2.5 leaked its native
+    # <|tool_calls_section_begin|> tokens as raw text where the template was
+    # missing a ToolCalls block); the k2.6 cloud build parses them correctly.
+    ResolutionRule(
+        family_substring="kimi",
+        driver="generic_native",
+        note="Native works on current Ollama; verified for kimi-k2.6:cloud",
     ),
     ResolutionRule(family_substring="llama", driver="generic_native"),
     ResolutionRule(family_substring="mistral", driver="generic_native"),

{ayder_cli-1.7.3 → ayder_cli-1.8.3}/src/ayder_cli/providers/impl/ollama_drivers/minimax.py

@@ -19,11 +19,11 @@ Available tools:
 </tools>
 
 To call a tool, wrap the call in <minimax:tool_call> tags using the
-function/parameter format:
+invoke/parameter format:
 <minimax:tool_call>
-<function=tool_name>
-<parameter=key>value</parameter>
-</function>
+<invoke name="tool_name">
+<parameter name="key">value</parameter>
+</invoke>
 </minimax:tool_call>
 """
 
@@ -33,7 +33,11 @@ class MiniMaxDriver(ChatDriver):
     mode = DriverMode.IN_CONTENT
     priority = 50
     fallback_driver = "generic_xml"
-    supports_families = ("minimax",)
+    # Dormant by default: the resolution matrix routes the minimax family to
+    # generic_native (Ollama parses tool calls natively). This IN_CONTENT driver
+    # stays available via explicit override / reactive fallback only, mirroring
+    # the dormant qwen3 and deepseek drivers.
+    supports_families = ()
 
     def render_tools_into_messages(
         self, messages: list[dict[str, Any]], tools: list[dict[str, Any]]