asky-cli 0.1.6__tar.gz

asky_cli-0.1.6/.gitignore

@@ -0,0 +1,12 @@
+repro_issue.py
+*.db
+temp_*
+temp/
+
+# Python packaging
+*.egg-info/
+dist/
+build/
+__pycache__/
+*.pyc
+.eggs/

asky_cli-0.1.6/DEVLOG.md

@@ -0,0 +1,237 @@
+# Development Log
+
+## 2026-02-02
+
+- **Fix**: Resolved **database bootstrapping failure** on first run.
+  - The CLI was attempting to count database records for the banner before the database was initialized.
+  - Moved `init_db()` to the beginning of `main()` in `cli.py`.
+  - Updated `tests/test_cli.py` to mock `get_db_record_count` in main flow tests to align with mocked `init_db`.
+
+## 2026-02-02
+
+- **Feat**: Enhanced CLI with a **shiny banner**.
+  - Renamed project from "asearch" to "asky".
+  - Added a cute icon.
+  - Added `get_banner` function to `banner.py` using `rich`.
+  - Integrated the banner into `cli.py` to show model details, summarizer details, search backend, and context limits in a rounded rectangle.
+  - **Expansion**: Reorganized the banner into a two-column layout to include **Default Model**, **Max Turns**, and **Database Record Count**.
+  - **Fix**: Adjusted banner styling to use `border_style="dim"` instead of a global `style="dim"`, ensuring icon and text colors remain vibrant.
+  - Ensured the banner is only shown for actual queries to maintain a clean interface for history and maintenance tasks.
+
+- **Feat**: Introduced **configurable query summarization threshold**.
+  - Added `continue_query_threshold` (default: 160) to `[general]` section in `config.toml`.
+  - Exposed `CONTINUE_QUERY_THRESHOLD` in `config.py`.
+  - Updated `generate_summaries` in `llm.py` to only summarize queries exceeding this threshold.
+  - Updated `get_interaction_context` in `storage.py` to use the full query if it's below the threshold, even if a summary exists.
+  - Added unit tests to verify the threshold logic.
+
+- **Feat**: Introduced **default context size setting** in the configuration.
+  - Added `default_context_size` (default: 4096) to `[general]` section in `config.toml`.
+  - Exposed `DEFAULT_CONTEXT_SIZE` in `config.py`.
+  - Updated `run_conversation_loop` in `llm.py` to use `DEFAULT_CONTEXT_SIZE` if not specified in model configuration.
+  - Added unit test in `test_config.py` to verify the setting.
+
+- **Feat**: Implemented **token usage tracking and reporting**.
+  - Added `UsageTracker` class to `llm.py` to accumulate token counts per model alias.
+  - Updated `get_llm_msg` to print the number of tokens sent in each turn.
+  - Enhanced `get_llm_msg` to extract real usage data from API responses if available (falls back to naive `len // 4` count).
+  - Updated `run_conversation_loop` and `generate_summaries` to report usage through the tracker.
+  - Added a session summary report at the end of the query execution, showing separate usage for the main model and the summarization model.
+  - Updated unit tests to support the new tracking logic.
+
+- **Fix**: Improved **rate limit handling and backoff robustness**.
+  - Increased `max_retries` from 5 to 10 in `llm.py`.
+  - Capped exponential backoff at 60 seconds to stay within API limit reset windows.
+  - Added support for respecting the `Retry-After` header from API responses (now handles decimal strings).
+  - Introduced a 1-second delay between sequential summarizations in `get_url_content` to reduce RPM (Requests Per Minute) pressure.
+  - Added unit tests to verify `Retry-After` handling and batch delay logic.
+- **Fix**: Resolved issue where **failed URL fetches were marked as "read"**.
+  - Updated `execute_get_url_details` in `tools.py` to only append to `read_urls` after a successful request.
+- **Fix**: Ensured `get_url_content` **respects the summarization flag** requested by the LLM.
+  - Previously, if the CLI was run without `-s`, the tool would ignore the LLM's own request for summaries and return full content.
+  - This often caused the next turn of the conversation to have a massive payload, triggering 429 rate limits on the LLM API itself.
+  - Now, `execute_get_url_content` uses `effective_summarize = args.get("summarize", summarize)`, correctly balancing user preference and LLM context management.
+- **Fix**: Added **URL sanitization** to remove shell-escaped backslashes.
+  - Wikipedia and other servers often return 403 Forbidden for URLs containing literal backslashes (e.g., `\(`, `\)`).
+  - These backslashes are frequently added when a user pastes a URL into a shell without proper quoting.
+  - Added `_sanitize_url` and applied it to `fetch_single_url` and `execute_get_url_details`.
+
+## 2026-02-01
+
+- **Feat**: Switched to `argparse.RawTextHelpFormatter` to allow explicit newlines in CLI help text.
+- **Feat**: Integrated **markdown rendering with `rich`**.
+
+  - Added `rich` to project dependencies.
+  - Implemented `is_markdown` detection in `llm.py`.
+  - Updated `run_conversation_loop` in `llm.py` and `print_answers` in `cli.py` to use `rich.markdown.Markdown` for rendering LLM output if markdown is detected.
+
+
+
+## 2026-02-01
+
+- **Refactor**: Consolidated `-f/--full` and `-s/--summarize` into a single `-s/--summarize` parameter.
+  - Removed `-f/--full` flag.
+  - Updated `-s/--summarize` to handle both URL content summarization and chat context summarization.
+  - Chat context now defaults to full content, and uses summaries only when `-s` is provided.
+  - Updated `cli.py` and consolidated tests in `test_cli.py`.
+
+
+## 2026-02-01
+
+- **Fix**: Added **configurable request timeout** for LLM API calls.
+  - Added `request_timeout` (default 60s) to `[general]` section in `config.toml`.
+  - Updated `get_llm_msg` in `llm.py` to use `REQUEST_TIMEOUT` from `config.py`.
+  - Broadened exception handling in `get_llm_msg` to catch and retry all `RequestException` errors (including timeouts).
+  - Verified with new unit test simulating connection timeouts and verifying retries.
+
+## 2026-02-01
+
+- **Refactor**: Removed **multi-threaded execution** from `get_url_content` tool.
+  - Replaced `ThreadPoolExecutor` with sequential execution to prevent rate limiting.
+  - Simplified `execute_get_url_content` logic.
+- **Feat**: Made **summarization prompts configurable** via `config.toml` under `[prompts]` section.
+  - Added `summarize_query` and `summarize_answer` templates with `{QUERY_SUMMARY_MAX_CHARS}` and `{ANSWER_SUMMARY_MAX_CHARS}` placeholders.
+  - Updated `llm.py` to use these templates for internal summarization tasks.
+
+- **Feat**: Made **User-Agent configurable** via `config.toml` under `[general]` section.
+  - Replaced hardcoded UA strings in `tools.py` with `USER_AGENT` constant.
+  - Introduced `llm_user_agent` for LLM requests in `llm.py`.
+  - Added support for loading both from configuration with sensible defaults.
+- **Fix**: Made `tests/test_cli.py` independent of local configuration by patching `DEFAULT_MODEL` and `MODELS`.
+
+- **Fix**: Added `User-Agent` header to all `requests` calls in `tools.py` to resolve **SearXNG 403 Forbidden** errors.
+- **Feat**: Introduced **custom user-defined tool support**.
+  - Users can define tools in `config.toml` under `[tool.NAME]`.
+  - Supports command execution via `subprocess`.
+  - Arguments are automatically quoted (inner quotes removed and wrapped in double quotes).
+  - Supports placeholder replacement (e.g., `command = "ls {path}"`) or positional appending.
+  - Default `list_dir` tool added to `config.toml`.
+- **Feat**: Added **clipboard support** via `/cp` slash command.
+  - Can be used anywhere in the query (not just at start).
+  - Integrates with `pyperclip`.
+- **Feat**: Improved **prompt expansion** to be recursive and work anywhere in the query string.
+- **Fix**: Made `tests/test_tools.py` independent of user's local configuration by explicitly patching `SEARCH_PROVIDER`.
+- **Refactor**: Cleaned up CLI code, removed unused imports and fixed lint warnings.
+- **Feat**: Introduced **Serper API** search provider support.
+  - Added `search_provider`, `serper_api_url`, and `serper_api_key_env` to `config.toml`.
+  - Refactored `execute_web_search` in `tools.py` to dispatch between `SearXNG` and `Serper`.
+  - Added full documentation comments for new configuration options.
+- **Fix**: Resolved `AttributeError: 'Namespace' object has no attribute 'prompts'` in CLI tests caused by recent changes.
+- **Refactor**: Moved configuration to `~/.config/asky/config.toml` (TOML format).
+- **Feat**: Added `ConfigLoader` to `src/asky/config.py` to auto-generate default config on first run.
+- **Feat**: Added support for defining API keys directly in `config.toml` or via environment variables (customizable names).
+- **Refactor**: Removed `python-dotenv` dependency and `.env` file loading logic.
+- **Refactor**: Relocated default history database to `~/.config/asky/history.db`. Path is configurable via `config.toml` or environment variable.
+- **Refactor**: Decoupled API and Model definitions in `config.toml`. Now supports `[api.name]` sections used by `[models.name]`.
+- **Feat**: Updated `config.py` to hydrate model configurations with API details (URLs, keys) automatically.
+- **Refactor**: Removed direct `LMSTUDIO` constant usage in favor of data-driven configuration.
+- **Test**: Verified new config schema with `pytest` and CLI execution (-v).
+- **Test**: Verified config generation, loading, and API key precedence. Ran regression tests (50 passed).
+- **Docs**: Added detailed explanatory comments to `src/asky/config.toml` explaining all sections and individual configuration fields.
+
+### Dependency Cleanup
+- Removed `liteLLM` dependency (simplified project, reduced bloat).
+- Implemented naive token counting: `len(content) // 4`.
+- Updated tests and matched `run_conversation_loop` tracking.
+
+### Debug Flag
+- Added `-v` / `--verbose` flag.
+- Prints full configuration (models, constants) at startup.
+- Prints LLM input previews (truncated) and tool usage status.
+
+### Database Cleanup & Bug Fixes
+- Added `--cleanup-db` (delete by ID/list/range/all)
+- Fixed duplicate history output (`ask -H`)
+- Fixed reverse ranges (e.g., `9-6`)
+- Restored `-fs` / `--force-search`
+
+### SQLite Conversation History
+- Local SQLite storage (queries, answers, model)
+- Added: `init_db`, `save_interaction`, `get_history`
+- Flags:
+  - `-H/--history` view recent history
+  - `-c/--continue` load context by ID
+  - `-f/--full` load full content
+- Numeric-only query → show past answers
+- Conversation loop returns final answer for storage
+- Added `force_search` logic to system prompt
+
+### Documentation
+- Updated `README.md` with "Key Features" and "How it Works" sections to provide a clearer overview of the tool's purpose and functionality.
+
+
+## 2026-01-31
+
+### Tools
+- `get_url_content` → fetch + strip HTML (4000 char cap, error handling)
+- `get_url_details` → content + links (Deep Dive, single-use)
+- `get_date_time` → ISO timestamp
+
+### HTML Processing
+- `HTMLStripper` (stdlib `html.parser`)
+- Skips `<script>/<style>`, extracts `<a>`
+- Major noise reduction (~734k → ~21k chars)
+
+### Tool Execution Engine
+- Multi-turn loop for sequential tool calls
+- Textual tool-call regex fixed (removed strict `$`)
+- Batch URL fetching + concurrency (`ThreadPoolExecutor`)
+- Single-use enforcement for heavy tools
+- Max turns increased to 15
+- Exponential backoff for `429` rate limits
+
+### CLI & Config
+- `argparse` integration
+- `-m/--model` selection
+- Unquoted queries supported (`nargs='+'`)
+- Per-model config (`max_chars`, auth, base_url)
+- `.env` support (`python-dotenv`, override enabled)
+- API auth via `api_key_env` or direct `api_key`
+- Added remote model examples (`gpt-4o`, Gemini)
+- Removed hardcoded keys
+
+### Modes
+- `-d/--deep-research [N]` → enforce multiple searches (default 3)
+- `-dd/--deep-dive` → recursive link exploration
+- Prompt construction moved to helpers/constants
+
+### Refactor & Cleanup
+- Fully modular; ≤30 lines per function
+- Type hints everywhere
+- Removed debug prints / unused imports
+- Safe import guard (`if __name__ == "__main__"`)
+- `strip_think_tags` removes `<think>` blocks
+
+### UX / Behavior
+- Query timing with `perf_counter`
+- Numeric-only queries auto-load history
+- Improved system prompts for tool usage
+- Direct SearXNG integration (removed `app.py` dependency)
+
+### Misc
+- Lint fixes
+- Improved errors and stability
+
+### Added Unit Tests
+- Installed `pytest` and configured it for the project.
+- Created comprehensive unit tests for core modules:
+    - `html.py`: HTML stripping and tag handling.
+    - `tools.py`: Web search, URL fetching, and date utilities (mocked network calls).
+    - `storage.py`: Database initialization, history storage, and cleanup (using temp DB).
+    - `config.py`: Configuration validation.
+    - `llm.py`: Helper functions and API call logic (mocked external APIs).
+- Verified all tests pass with `uv run pytest`.
+
+
+### Expanded Unit Coverage
+- Created `tests/test_cli.py` to cover all CLI functionality including argument parsing, history display, and context loading.
+- Updated `tests/test_llm.py` to test the core conversation loop (`run_conversation_loop`) and summarization logic (`generate_summaries`) using mocks.
+- Updated `tests/test_tools.py` to test text summarization (`summarize_text`).
+- Updated `tests/test_storage.py` to cover edge cases in database cleanup (e.g., reverse ranges).
+- Achieved 100% pass rate for 47 unit tests.
+- **Refactor**: Replaced dynamic configuration generation with a bundled `config.toml`. The default configuration file is now shipped with the package and copied to the user's config directory on first run.
+- **Feat**: Enhanced `asky -c` (continue) to support relative history IDs. Users can now use `~1` (most recent), `~2` (second most recent), etc., instead of looking up exact database IDs.
+- **Feat**: Added support for **Predefined Prompts**. Users can define reusable prompts in `config.toml` under `[user_prompts]` and invoke them via `/key` (e.g., `ask /wh Rotterdam`).
+- **Feat**: Added `--prompts` / `-p` flag to list configured user prompts.
+- **Change**: Renamed `--print-answer` short flag from `-p` to `-pa` to accommodate the new prompts flag.
+

asky_cli-0.1.6/PKG-INFO

@@ -0,0 +1,290 @@
+Metadata-Version: 2.4
+Name: asky-cli
+Version: 0.1.6
+Summary: AI-powered web search CLI with LLM tool-calling capabilities
+License-Expression: MIT
+Requires-Python: >=3.10
+Requires-Dist: pyperclip>=1.11.0
+Requires-Dist: requests
+Requires-Dist: rich
+Description-Content-Type: text/markdown
+
+
+<img src="assets/asearch_icon.png" alt="asky icon" width="200" align="right">
+
+<font size="6">**asky**</font>
+
+
+AI-powered web search CLI with LLM tool-calling capabilities.
+
+asky (can be invoked as `asky` or `ask`) is a powerful command-line interface that brings AI-powered search and research capabilities directly to your terminal. It uses LLMs and tools to synthesize answers from the web (or from files and apps you have on your computer).
+## Key Features
+
+- **Multi-Model Support**: Easily define and switch between various LLMs and providers that supports OpenAI compatible API.
+- **Tool-Calling Integration**: Models can autonomously perform web searches (via SearXNG or Serper API), fetch URL content, and get current date/time to provide accurate, up-to-date answers.
+- **Intelligent Content Fetching**: Automatically strips HTML noise (scripts, styles) to provide clean text context to the models.
+- **Conversation History**: Maintains a local SQLite database of your queries and answers, allowing for context-aware follow-up questions.
+- **Deep Research Mode**: Automatically performs multiple searches to provide comprehensive analysis of complex topics.
+- **Deep Dive Mode**: Recursively explores links found on web pages for in-depth information gathering.
+- **Predefined Prompts**: Save and quickly invoke common prompt patterns using simple slashes (e.g., `/gn` for get latest news from The Guardian).
+- **Clipboard Integration**: Use `/cp` to expand the query with clipboard content.
+- **Custom Tools**: Expose any CLI command as a tool for the LLM. Define your own commands and parameters in `config.toml`.
+
+## How it Works
+
+1. **User Query**: You provide a query to the `ask` command.
+2. **Model Selection**: asky initializes the selected LLM based on your configuration.
+3. **Tool Loop**: The LLM analyzes your query. If it needs real-world data, it calls integrated tools (like `web_search`).
+4. **Context Synthesis**: asky fetches the data, cleans it, and feeds it back to the LLM. This process can repeat for up to 15 turns for complex research.
+5. **Final Answer**: The LLM synthesizes all gathered information into a concise, formatted response.
+6. **Persistence**: The interaction is saved to your local history for future reference.
+
+
+## Installation
+
+```bash
+pip install ask-cli
+```
+
+Or install from source:
+
+```bash
+pip install -e .
+```
+
+## Usage
+
+```
+
+# Basic query
+ask what is the weather in Berlin
+
+# Continue from previous query (by ID)
+ask -c 1 tell me more about that
+
+# Continue from last query (relative ID)
+ask -c~1 explain more
+# OR
+ask -c "~2" what about the one before that?
+
+> [!NOTE]
+> **Zsh Users**: When using `~` for relative IDs, you must either quote the value (e.g., `ask -c "~1"`) or place it immediately after the flag without a space (e.g., `ask -c~1`). If you use a space without quotes (e.g., `ask -c ~1`), zsh will attempt to expand it as a directory stack entry.
+
+
+➜  ~ ask -p
+
+=== USER PROMPTS ===
+  /gn         : Give me latest news from The Guardian, use https://www.theguardian.com/europe
+  /wh         : how is weather in
+====================
+
+➜  ~ ask /wh delft
+Dispatching tool call: web_search with args {'q': 'weather in Delft'}
+Dispatching tool call: get_url_content with args {'urls': ...}
+
+The weather in **Delft, South Holland, Netherlands** is currently **45°F and Cloudy with Showers in the Vicinity** (as of 4:20 pm CET).
+
+Here is the forecast for today and the next couple of days:
+
+...
+
+Query completed in 3.88 seconds
+
+--------------------------------------------------------------------------------
+➜  ~ ask --help
+usage: ask [-h] [-m {gf,glmair,glmflash,q34t,q34,lfm,q8,q30,onano,omini}] [-d [DEEP_RESEARCH]] [-dd] [-c CONTINUE_IDS] [-s] [-fs] [--cleanup-db [CLEANUP_DB]] [--all]
+           [-H [HISTORY]] [-pa PRINT_IDS] [-p] [-v]
+           [query ...]
+
+Tool-calling CLI with model selection.
+
+positional arguments:
+  query                 The query string
+
+options:
+  -h, --help            show this help message and exit
+  -m, --model {gf,glmair,glmflash,q34t,q34,lfm,q8,q30,onano,omini}
+                        Select the model alias
+  -d, --deep-research [DEEP_RESEARCH]
+                        Enable deep research mode (optional: specify min number of queries, default 5)
+  -dd, --deep-dive      Enable deep dive mode (extracts links and encourages reading more pages from same domain)
+  -c, --continue-chat CONTINUE_IDS
+                        Continue conversation with context from specific history IDs (comma-separated, e.g. '1,2').
+  -s, --summarize       Enable summarize mode (summarizes URL content and uses summaries for chat context)
+  -fs, --force-search   Force the model to use web search (default: False).
+                        Helpful for avoiding hallucinations with small models
+  --cleanup-db [CLEANUP_DB]
+                        Delete history records. usage: --cleanup-db [ID|ID-ID|ID,ID] or --cleanup-db --all
+  --all                 Used with --cleanup-db to delete ALL history.
+  -H, --history [HISTORY]
+                        Show last N queries and answer summaries (default 10).
+                        Use with --print-answer to print the full answer(s).
+  -pa, --print-answer PRINT_IDS
+                        Print the answer(s) for specific history IDs (comma-separated).
+  -p, --prompts         List all configured user prompts.
+  -v, --verbose         Enable verbose output (prints config and LLM inputs).
+```
+
+**Deep research mode** (encourages model to perform multiple searches)
+
+ask -d 5 comprehensive analysis of topic
+
+**Deep dive mode** (encourages model to read multiple pages from same domain)
+
+ask -dd https://example.com
+
+**Use a specific model**
+
+ask -m gf what is quantum computing
+
+**Force web search**
+
+ask -fs latest news on topic
+
+
+**Pre-configured model definitions**
+
+Followin model definitions ship with default config.toml, but you can add any number of models that are served with an OpenAI compatible API.
+
+- `gf` - Google Gemini Flash (default)
+- `lfm` - Liquid LFM 2.5
+- `q8` - Qwen3 8B
+- `q30` - Qwen3 30B
+- `q34` - Qwen3 4B
+- `q34t` - Qwen3 4B Thinking
+
+## Custom Tools
+
+You can define your own tools in `config.toml` that the LLM can use to interact with your local system. Each tool runs a CLI command and returns the output to the LLM.
+
+Example configuration for a `list_dir` tool:
+
+```toml
+[tool.list_dir]
+command = "ls"
+description = "List the contents of a directory."
+
+[tool.list_dir.parameters]
+type = "object"
+required = ["path"]
+
+[tool.list_dir.parameters.properties.path]
+type = "string"
+default = "."
+```
+
+Example configuration for a `grep_search` tool:
+
+```toml
+[tool.grep_search]
+command = "grep -r {pattern} {path}"
+description = "Search for a pattern in files recursively."
+
+[tool.grep_search.parameters]
+type = "object"
+required = ["pattern"]
+
+[tool.grep_search.parameters.properties.pattern]
+type = "string"
+description = "The regex pattern to search for."
+
+[tool.grep_search.parameters.properties.path]
+type = "string"
+description = "The directory path to search in."
+default = "."
+```
+
+> [!CAUTION]
+> **Security Risk**: Custom tools execute commands using your system shell. While asky attempts to quote arguments safely, exposing powerful CLI tools to an LLM can be risky. Use this feature with caution.
+
+### How it works:
+- **Placeholders**: Use `{param_name}` in the `command` string to inject arguments. If no placeholders are found, arguments are appended to the command.
+- **Quoting**: All arguments are automatically cleaned (inner double-quotes removed) and wrapped in double-quotes for safety.
+- **Execution**: Commands are executed via terminal shell, allowing for advanced piping and redirection.
+
+> [!TIP]
+> **Performance Tip**: When using recursive tools like `grep`, consider excluding large directories like `.venv` or `node_modules` to avoid timeouts:
+> `command = "grep -r --exclude-dir={.venv,node_modules} {pattern} {path}"`
+
+> [!NOTE]
+> **Optional Parameters**: If you define a parameter with a `default` value in `config.toml`, it will be automatically injected into your `command` if the LLM omits it.
+
+## Configuration options
+[See default configuration](./src/asky/config.toml)
+
+
+On first run, a default configuration file is created at `~/.config/asky/config.toml`. You can edit this file to configure models, API keys, and other settings.
+
+### API Keys
+You can set API keys in two ways:
+1. **Environment Variables**: Set `GOOGLE_API_KEY` (or other configured env vars) in your shell.
+2. **Config File**: Add keys directly to `[api.name]` sections in `config.toml`.
+
+Example `config.toml`:
+```toml
+[general]
+default_model = "gf"
+
+[api.gemini]
+api_key_env = "GOOGLE_API_KEY"
+
+[api.lmstudio]
+url = "http://localhost:1234/v1/chat/completions"
+```
+
+### Verification
+Run with `-v` to see the loaded configuration:
+```bash
+ask -v
+```
+
+
+## Web Search
+
+asky works best with a web search tool. You can use SearXNG or Serper API. 
+
+### Serper API
+Serper is a paid service, but gives 2500 requests for free. 
+
+### Install & configure SearXNG
+SearXNG is free and open source, it's easy to set up with a single docker command.
+
+Following command taken from [SearXNG docs](https://docs.searxng.org/admin/installation-docker.html#instancing).
+```bash
+docker pull docker.io/searxng/searxng:latest
+
+# Create directories for configuration and persistent data
+$ mkdir -p ./searxng/config/ ./searxng/data/
+$ cd ./searxng/
+
+# Run the container
+$ docker run --name searxng -d \
+    -p 8888:8080 \
+    -v "./config/:/etc/searxng/" \
+    -v "./data/:/var/cache/searxng/" \
+    docker.io/searxng/searxng:latest
+```
+You need to add "-json" to the formats section of the default searxng config.yaml file.
+```yaml
+  # remove format to deny access, use lower case.
+  # formats: [html, csv, json, rss]
+  formats:
+    - html
+    - json
+```
+Then restart the container.
+```bash
+docker restart searxng
+```
+
+
+
+## Requirements
+
+- Python 3.10+
+- Running SearXNG instance or Serper API key.
+- LM Studio (for local models) or API keys for remote models
+
+## License
+
+MIT