# PR Description
The `arcade/pyproject.toml` wasn't able to find the `README.md` file
because it must be a subpath of `arcade-ai/arcade`. I created a simple
README for PyPi
# PR Description
Given that `arcade chat` is the entry point for all users, this PR makes
the developer's first experience with Arcade a better experience.
1. Add multi-line & pasting support to `arcade chat`. To start a
multi-line chat message, start your message with `"""`. To end the
multi-line chat message, end your line with `"""`.
2. Added chat commands
- `/?` displays all available chat commands that are available
- `/show` shows all of the tools available. This uses `arcade show`.
Currently no option to specify the toolkit, or tool.
- `/clear` clears the chat conversation history of the current session
- `/exit` exits the chat. (ctrl+c still works of course!)
1. Fixes bug where arcade login doesn't work for localhost
- `arcade login -h localhost` will open login page at
`http://localhost:8000/...`
- Optionally specify the port: `arcade login -h localhost -p 8000`
3. Adds `local` flag to `arcade show`
- `-h localhost`, `-h 127.0.0.1`, and `-h 0.0.0.0` shows the tools that
are in the local engine's catalog
- `--local` show the tools that are in the local environment.
Fixes an edge case where the actor doesn't start because no toolkits are
installed in the local environment, but `arcade dev` keeps waiting for a
healthy actor.
Context: Currently, `arcade dev` starts the actor process and then waits
a hardcoded amount of time (2sec) for the actor to start up. This isn't
enough time on some slower machines, which leads to the engine trying to
start but failing.
Fix: Wait until the actor is healthy according to its own
`/actor/health` endpoint.
I ran into this and was scratching my head why `arcade login` was
exiting with no output.
Fix: Get config data from the new top-level object we introduced.
Symptom: If an `otel_meter` wasnot passed when constructing the actor
(which is allowed, it's an optional param), the actor would later crash
at runtime.
Fix: Always set `tool_counter` to a default value.
Fixes a circular import issue where `arcade.sdk -> arcade.core` but also
`arcade.core -> arcade.sdk`. My mistake!
Moved some of the shared classes down into `core`, and re-exported them
to `sdk` to keep the expected interface for devs.
This PR ensures that `arcade.core` does not show up anywhere in "user
space". This is crucial for helping developers understand what objects
are safe to use, and helps maintain a good developer experience.
Specific changes:
- `ToolAuthorizationContext` and `ToolContext` are now visible via
`arcade.sdk`
- `ToolCatalog` is now visible via `arcade.sdk`
- `Toolkit` is now visible via `arcade.sdk`
- `config` is now visible via `arcade.sdk.config`
# PR Description
1. `arcade show` only supported lowercase args for toolkit and tools.
This PR allows the user to use capitalization and the tool/toolkit will
still be displayed.
2. `arcade show -T` is now for showing a toolkit and `arcade show -t` is
now for showing a tool.
fix for version:
Dynamic Version Retrieval: Replaces the hardcoded version import with
dynamic retrieval using importlib.metadata.version. This ensures that
__version__ in __init__.py accurately reflects the installed version of
arcade-ai.
fix to ``arcade-new``: Changes the default dependency specification for
arcade-ai from ^{VERSION} to ~{VERSION} in DEFAULT_VERSIONS in
``new.py``. This adjustment allows for patch version updates while
maintaining compatibility, enhancing stability.
---------
Co-authored-by: Eric Gustin <eric@arcade-ai.com>
# PR Description
This PR improves the CLI command `arcade new` in multiple ways.
1. When prompting the developer if they want to create a test/evals
folder, the default was printed twice. This is now fixed.
2. Whether to generate a test/eval directory, the user must enter "yes".
This PR relaxes this to accept other variations.
3. Creates an eval suite instead of an empty folder.
4. Prevents a developer from overwriting an existing toolkit.
**New Tools Added**
- `docs.py`: Provides tools for Google Docs functionalities, including
creating documents and inserting text.
- `drive.py`: Introduces tools for Google Drive operations, such as
listing documents.
This PR also focuses on simplifying the error handling logic in the Google
toolkit, specifically within the Calendar and Gmail tools. The primary
change involves removing redundant `try-except` blocks that were
catching `HttpError` and general exceptions, and re-raising them as
`ToolExecutionError`. By removing these blocks, we allow exceptions to
propagate naturally, and be handled by the ``ToolExecutor``
---------
Co-authored-by: Eric Gustin <eric@arcade-ai.com>
This PR introduces the following changes:
- **Engine Environment Configuration**: Adds support for specifying an
environment variables file for the engine via the `arcade dev` CLI
command.
- **Configuration File Handling**: Refactors configuration file handling
in the CLI launcher to generalize logic for locating configuration
files.
- **Tool Execution Logging**: Enhances logging in `BaseActor` to include
execution duration and adjusts logging levels for better visibility.
- **Enhanced Tool Exception Handling**: Improves exception handling in
`ToolExecutor` and updates the `@tool` decorator to ensure proper
propagation and handling of exceptions raised during tool execution.
# PR Description
When a tool call required authorization, `arcade chat` would hit rate
limits extremely quickly when waiting to authorize with a given link.
This PR introduces using long polling when sending a GET request to
Arcade API `/auth/status`. The `/auth/status` endpoint supports the
`wait` query parameter, where if present, will not respond until either
auth status becomes `completed` or `timeout` is reached. If the
`timeout` is reached, `arcade chat` catches the 408 response and tries
again. For `arcade chat` we set the `wait` query param to 60 seconds.
## Problem
I found a bug with `Github.CountStargazers` where a stargazer count of
`0` was interpreted as a null result. In other words, 0 wasn't passed
back to the Engine correctly.
Separately, the tool function was also not authorized correctly.
## Fix
- Don't use a falsy comparison when evaluating `result` inside the
`ToolOutputFactory`
- Add unit tests for `ToolOutputFactory` to give us confidence in the
business logic
- Added `ToolContext` to pass in the authorization token correctly.
Before
```
User (nate@arcade-ai.com):
how many stars does the ArcadeAI/Docs repo have on github?
Assistant (gpt-4o):
I successfully checked the repository, but unfortunately, I cannot provide the number of stars for the ArcadeAI/Docs repository. Please try checking directly on GitHub for the most accurate information.
Called tool 'Github_CountStargazers'
Parameters:{"owner":"ArcadeAI","name":"Docs"}
'Github_CountStargazers' tool returned:Github.CountStargazers called successfully
```
After
```
User (nate@arcade-ai.com):
how many stars does the ArcadeAI/Docs repo have on github?
Assistant (gpt-4o):
The ArcadeAI/Docs repository on GitHub has 0 stars.
Called tool 'Github_CountStargazers'
Parameters:{"owner":"ArcadeAI","name":"Docs"}
'Github_CountStargazers' tool returned:0
This PR introduces enhancements and fixes to the Evaluation Framework to
improve accuracy, robustness, and performance.
## Key Improvements
- **Refactored Evaluation Loading**: Extracted evaluation file loading
and suite loading logic into utility functions `get_eval_files` and
`load_eval_suites` in `arcade/cli/utils.py`.
- **Benefit**: Enhances code modularity and maintainability.
- **Asynchronous Execution of Evaluations**: Modified the evaluations to
run asynchronously using `asyncio`.
- **Benefit**: Significantly reduces total execution time when running
multiple evaluations.
- **Improved Error Handling**: Wrapped critic evaluations in try-except
blocks to handle exceptions gracefully.
- **Benefit**: Ensures that a single failing critic doesn't halt the
entire evaluation process.
## Other Changes
- **Case-Insensitive Tool Name Comparison**: Made tool name comparisons
case-insensitive to improve robustness against casing differences.
- **Refactored Cost Matrix Creation**: Revised cost matrix creation to
handle varying numbers of expected and actual tool calls properly,
ensuring accurate assignment and scoring.
- **Type Casting in `BinaryCritic`**: Added type casting for actual
values to match the expected value's type before comparison, improving
accuracy in evaluations.
- **Removed Synchronous Code Paths**: Simplified the codebase by
removing synchronous evaluation methods, focusing on asynchronous
execution.
- **General Code Cleanup**: Removed unused imports and performed general
code cleanup to enhance readability and maintainability.
## Example
For the google calendar and gmail tools eval set you can run
```go
> arcade evals . -c 8 --models gpt-4o,gpt-4o-mini,gpt-4,gpt-4-turbo
Running evaluations in calendar_eval_suite
Running evaluations in gmail_eval_suite
Model: gpt-4o
PASSED Create calendar event -- Score: 1.00
FAILED List calendar events -- Score: 0.86
PASSED Update a calendar event -- Score: 1.00
PASSED Delete a calendar event -- Score: 1.00
Model: gpt-4o-mini
FAILED Create calendar event -- Score: 0.84
FAILED List calendar events -- Score: 0.86
PASSED Update a calendar event -- Score: 1.00
PASSED Delete a calendar event -- Score: 1.00
Model: gpt-4
PASSED Create calendar event -- Score: 1.00
FAILED List calendar events -- Score: 0.86
PASSED Update a calendar event -- Score: 1.00
PASSED Delete a calendar event -- Score: 1.00
Model: gpt-4-turbo
PASSED Create calendar event -- Score: 1.00
FAILED List calendar events -- Score: 0.87
PASSED Update a calendar event -- Score: 1.00
PASSED Delete a calendar event -- Score: 1.00
Model: gpt-4o
PASSED Send email to user with clear username -- Score: 1.00
Model: gpt-4o-mini
PASSED Send email to user with clear username -- Score: 1.00
Model: gpt-4
PASSED Send email to user with clear username -- Score: 1.00
Model: gpt-4-turbo
PASSED Send email to user with clear username -- Score: 1.00
Summary -- Total: 20 -- Passed: 15 -- Failed: 5
```
```
authors = ["Arcade AI <dev@arcade-ai.com>"]
```
vs
```
authors = ["Arcade AI <dev@arcade-ai.com"]
```
There is also now a ``make`` command for ``make install-toolkits``
Currently with arcade chat, when we send a request that requires tool
authentication, it sends the message then, waits for authentication,
then resends the current chat history. However we resend the
authntication link as the latest message instead of resending the
message that prompted the authentication link.
OpenAI ignores this and continues, however Anthropic errors because it
sees the history as
-User message {Get me sams last 5 tweets}
-Assistant message {Please go to this URL and authorize the action: }
instead of
-User message {Get me sams last 5 tweets}
-Assistant message {Please go to this URL and authorize the action: }
-User message {Get me sams last 5 tweets}
Having an assistant message as the latest message in a request is
invalid for anthropic
### Adds the following tools to the Github Toolkit:
1. CreateIssueComment
2. SetStarred
3. CountStargazers
4. ListOrgRepositories
5. GetRepository
6. ListRepositoryActivities
7. ListReviewCommentsInARepository
8. ListPullRequests
9. GetPullRequest
10. UpdatePullRequest
11. ListPullRequestCommits
12. CreateReplyForReviewComment
13. ListReviewCommentsOnPullRequest
14. CreateReviewComment
Adds evals for all of these tools and unit tests.
---------
Co-authored-by: Sam Partee <sam@arcade-ai.com>
This PR adds four new tools to the Google ToolKit
* `create_event`
* `list_events`
* `update_event`
* `delete_event`
I also improved an error log when tools are being registered by the
actor.
This PR also sneaks in an eval for gmail
Here is a sample conversation that shows the tools and their
capabilities and limitiations:
This PR improves the Docker build process by shifting from building the
project within the Docker image to using pre-built wheels. The main
changes are:
1. **Updated Makefile:**
- **`VERSION` Variable:** Set to `0.1.0.dev0` to reflect the new default
development version.
- **`docker` Target:**
- Added steps to build the Arcade and toolkit wheels before building the
Docker image.
- Exports the required extras (`fastapi`, `evals`) to a
`requirements.txt` file.
- **`full-dist` Target:**
- Builds distributions for the main project and all toolkits.
- Copies all the built wheels to a centralized `./dist` directory.
- **`clean-dist` Target:**
- Cleans build artifacts from `./dist`, `arcade/dist`, and
`toolkits/*/dist` directories.
2. **Modified Dockerfile:**
- **Copy Pre-built Wheels:** Adjusted to copy wheels and the
`requirements.txt` from the `./dist` directory into the Docker image.
- **Installation Process:**
- Installs the Arcade wheel with the necessary extras.
- Installs toolkits from the copied wheel files, eliminating the need to
build them inside the Docker image.
- **Simplification:** Removed unnecessary commands, such as installing
build tools and copying the entire codebase, to streamline the
Dockerfile.
3. **Toolkits `pyproject.toml` Updates:**
- Changed the `arcade-ai` dependency version from `^0.1.0` to `0.1.*` in
all toolkit `pyproject.toml` files to ensure compatibility with the new
versioning scheme.
4. **Docker Makefile Adjustments:**
- Set the `VERSION` variable to `0.1.0.dev0` to align with the main
Makefile.
- Ensures consistent versioning across Docker-related build processes.
**Benefits:**
- **Efficiency:** Building wheels outside the Docker context reduces the
Docker image build time and resource consumption. overall docker image
size reduced by **1Gb**!!!
- **Reliability:** Using pre-built wheels ensures consistency across
different environments and simplifies dependency management.
- **Maintainability:** The Dockerfile and Makefiles are cleaner and more
straightforward, making them easier to understand and maintain.
**Notes:**
- Developers should run `make docker` to build and run the Docker
container using the new process.
- Ensure that any CI/CD pipelines are updated to accommodate these
changes in the build process. @sdreyer
On the last few PRs I have noticed two problems:
1. `ruff format` fails even though it seems OK on our local machines
(sometimes, not always)
2. Nate's and Sam's machines kept flip-flopping a specific piece of
formatting back and forth, indicating a subtle difference of config
hiding somewhere
3. This was reproducible by running `ruff format` in the terminal,
followed by `make check`. The former would edit files, and then `make
check` would edit them back!
This PR addresses both issues, and further standardizes our editor &
linter configs to be super stable.
Specifically:
1. The main fix for the above, the pre-commit hook was pinned to a super
old version of ruff.
This resulted in subtle differences in behavior between our machines,
and on CI.
2. Moved ruff settings from `pyproject.toml` to `.ruff.toml`
pyproject files in subdirectories (e.g. `toolkits/**`) were overriding
the main pyproject file and erasing the custom ruff config we set at the
root. This meant that our ruff config was applied to `arcade` but not to
any of the other packages.
By moving the config to `.ruff.toml` at the root, all projects will
inherit the same ruff linting & formatting config.
4. Un-ignored the `.vscode/` directory so that we can share
vscode/cursor workspace settings.
This is valuable for standardizing settings like the default formatter
(ruff) and default test framework (pytest).
However, it's important that going forward we _only_ commit things here
that should apply across all of our machines.
5. To avoid any conflict between prettier and ruff, prettier now
explicitly ignores *.py files
6. Finally, `ruff format` and `make check` agree. A number of files are
newly auto-formatted.
# Address Alpha Release Tweaks and Bugs
This PR addresses several issues and tweaks identified during the alpha
release:
- **Ensure `~/.arcade` directory exists before writing the config file**
In `arcade/cli/authn.py`, added code to create the `~/.arcade` directory
if it doesn't exist. This prevents errors when writing the configuration
file during the login process.
- **Fix retry logic in process management**
In `arcade/cli/launcher.py`, corrected an off-by-one error in the retry
logic within the `_manage_processes` function. This ensures that the
process management behaves as expected when retries are exhausted.
- **Allow passing environment variables to the engine process**
(technically this option isn't exposed yet)
Updated the `start_servers`, `_manage_processes`, and `_start_process`
functions in `arcade/cli/launcher.py` to accept an `engine_env`
parameter. This allows custom environment variables to be set for the
engine process. Also, set `GIN_MODE` to `"release"` by default.
- **Handle cases with no critics in evaluations**
Modified the `EvalCase` class in `arcade/sdk/eval/eval.py` to handle
scenarios where no critics are provided. This avoids potential errors
during the evaluation process when critics are absent. Should add a test
for this.
- **Adjust dependencies in `pyproject.toml`**
- Moved `uvicorn` to be an optional dependency and included it in the
`fastapi` extra.
- Removed unnecessary development dependencies (`mkdocs`,
`mkdocs-material`, `mkdocstrings`).
- Ensured that `uvicorn` is updated to version `^0.30.0`.
---------
Co-authored-by: Nate Barbettini <nate@arcade-ai.com>
