AutoGPT is the vision of accessible AI for everyone, to use and to build on. Our mission is to provide the tools, so that you can focus on what matters.
MIT License
Bot releases are visible (Hide)
New OpenAI Models π€
gpt-4-turbo
is now supported and set as the default SMART_LLM
model.FAST_LLM
changed from gpt-3.5-turbo-16k
to gpt-3.5-turbo
.EMBEDDING_MODEL
changed from text-embedding-ada-002
to text-embedding-3-small
.gpt-4-0125-preview
and gpt-4-turbo
models.Agent Protocol Server Enhancements π
AP_SERVER_PORT
environment variable.AgentProtocolServer
.CLI Usability Improvements π»
Web Browsing Enhancements π
extract_information
function to extract pieces of information from webpage content based on a list of topics of interest. The read_webpage
command now supports topics_of_interest
and get_raw_content
parameters to leverage this capability.Telemetry & Error Tracking π
production
and dev
environments based on VCS state.File Storage Abstraction π
FileStorage
class.FileWorkspace
to FileStorage
and updated associated classes and methods.AgentManager
and AgentProtocolServer
to use the new FileStorage
system.History Compression π
JSON Parsing Robustness
json_loads
for more tolerant JSON parsing of llm responses.extract_dict_from_response
to handle both json and
JSON blocks in responses.extract_dict_from_response
.Error Handling
action_history
related exceptions in CLI and Server modes.Artifact & File Handling
agent_created
attribute instead of registering a new Artifact.read_file
command in GCS and S3 workspaces.Dependency Updates & Security Fixes
aiohttp
and fastapi
dependencies to mitigate vulnerabilities.python-multipart
.TTY Mode Enhancements
finish
command behavior in TTY mode.AgentTerminated
exception to exit the loop.Miscellaneous Fixes
summarize_text
and QueryLanguageModel
abilities to handle OpenAI API changes.auto-gpt-plugin-template
to regular dependencies to fix missing module error.agbenchmark
and autogpt-forge
dependencies across the project.test_gcs_file_workspace
by changing the fixture scope.Note: most of the changes mentioned above were made through direct commits. See also the full changelog.
agbenchmark
to directory dependency in autogpt
and forge
by @Pwuts in https://github.com/Significant-Gravitas/AutoGPT/pull/6946
action_history
-related exceptions gracefully by @kcze in https://github.com/Significant-Gravitas/AutoGPT/pull/6990
polyfill.io
by @SukkaW in https://github.com/Significant-Gravitas/AutoGPT/pull/6952
json_loads
function by @kcze in https://github.com/Significant-Gravitas/AutoGPT/pull/7016
pyyaml
loader with SafeLoader
by @matheudev in https://github.com/Significant-Gravitas/AutoGPT/pull/7035
web_search.py
to duckduckgo_search
v5 by @kcze in https://github.com/Significant-Gravitas/AutoGPT/pull/7045
save_state
behave like save as by @kcze in https://github.com/Significant-Gravitas/AutoGPT/pull/7025
create_chat_completion
by @Pwuts in https://github.com/Significant-Gravitas/AutoGPT/pull/7082
Full Changelog: https://github.com/Significant-Gravitas/AutoGPT/compare/autogpt-v0.5.0...autogpt-v0.5.1
Published by Pwuts 10 months ago
First some important notes w.r.t. using the application:
run.sh
has been renamed to autogpt.sh
autogpts/autogpt
.autogpt.sh serve
in the subproject folder (autogpts/autogpt
) will also serve the new frontend, but without benchmarking functionality.autogpt.sh run
..env.template
.latest
, v0.5.0
in this case), we now also publish a latest-dev
image that always contains the latest working build from master
. This allows you to try out the latest bleeding edge version, but be aware that these builds may contain bugs!PromptStrategy
To make it easier to harness the power of LLMs and use them to fulfil tasks within the application, we adopted the PromptStrategy
class from autogpt.core
(AKA re-arch) to encapsulate prompt generation and response parsing throughout the application.
Config modularization
To reduce the complexity of the application's config structure, parts of the monolithic Config
have been moved into smaller, tightly scoped config objects. Also, the logic for building the configuration from environment variables was decentralized to make it all a lot more maintainable.
This is mostly made possible by the autogpt.core.configuration
module, which was also expanded with a few new features for it. Most notably, the new from_env
attribute on the UserConfigurable
field decorator and corresponding logic in SystemConfiguration.from_env()
and related functions.
Monorepo
As mentioned, the repo has been restructured to accommodate the AutoGPT Agent, Forge, AGBenchmark and the new Frontend.
autogpts/autogpt
autogpts/forge
, and the project's new CLI makes it easy to create new Forge-based agents.benchmark
frontend
See also the README.
Note: most of the changes mentioned above were made through direct commits. See also the full changelog.
execute_python_file
by @MauroDruwel in https://github.com/Significant-Gravitas/AutoGPT/pull/3972
core
by @Pwuts in https://github.com/Significant-Gravitas/AutoGPT/pull/5286
Config
and revive Azure support by @Pwuts in https://github.com/Significant-Gravitas/AutoGPT/pull/6497
duckduckgo-search
version to v4.0.0 to unbreak web_search
command by @zedatrix in Published by github-actions[bot] about 1 year ago
Published by lc0rp about 1 year ago
Auto-GPT v0.4.7 introduces initial REST API support, powered by e2b's agent protocol SDK. It also includes improvements to prompt generation and support for our new benchmarking tool, Auto-GPT-Benchmarks.
We've also moved our documentation to Material Theme at https://docs.agpt.co. And, as usual, we've squashed a few bugs and made under-the-hood improvements.
master
by @Pwuts in https://github.com/Significant-Gravitas/Auto-GPT/pull/5065
Full Changelog: https://github.com/Significant-Gravitas/Auto-GPT/compare/v0.4.6...v0.4.7
Published by Pwuts about 1 year ago
plugin.handle_text_embedding
hook by @zhanglei1172 in https://github.com/Significant-Gravitas/Auto-GPT/pull/2804
execute_python_file
workspace mount & Windows path formatting by @sohrabsaran in https://github.com/Significant-Gravitas/Auto-GPT/pull/4996
Full Changelog: https://github.com/Significant-Gravitas/Auto-GPT/compare/v0.4.5...v0.4.6
Published by lc0rp over 1 year ago
This maintenance release includes under-the-hood improvements and bug fixes, such as more accurate token counts for OpenAI functions, faster CI builds, improved plugin handling, and refactoring of the Config class for better maintainability.
We have released some documentation updates, including:
How to share system logs
Auto-GPT re-architecture documentation
MessageHistory
on ChatSequence
by @Pwuts in https://github.com/Significant-Gravitas/Auto-GPT/pull/4922
Config.plugins
as AutoGPTPluginTemplate
by @Pwuts in https://github.com/Significant-Gravitas/Auto-GPT/pull/4924
stable
back into master
by @Pwuts in https://github.com/Significant-Gravitas/Auto-GPT/pull/4947
Full Changelog: https://github.com/Significant-Gravitas/Auto-GPT/compare/v0.4.4...v0.4.5
Published by lc0rp over 1 year ago
Auto-GPT v0.4.4 is dedicated to the core re-arch tram, led by @collijk.
This release is noteworthy for two reasons.
Firstly, it comes hot on the heels of OpenAI's GA release of GPT-4. Auto-GPT users have eagerly awaited the opportunity to unlock more power via a GPT-4 model pairing. In v0.4.4, the SMART_LLM (formerly SMART_LLM_MODEL) defaults to GPT-4 once again, and we have implemented adjustments to ensure the correct usage of SMART_LLM and FAST_LLM (formerly FAST_LLM_MODEL) throughout the code-base. The smarter option is used consistently for areas requiring state-of-the-art accuracy, such as agent command selection. At the same time, the faster LLM assists with tasks that even the speedier GPT-3.5-turbo excels at, like summarization.
Note: GPT-4 is costlier, so please review your SMART_* and FAST_* settings. You can also use --gpt3only and --gpt4only command line flags to adjust your model preferences at runtime.
The second reason, and the reason for the dedication at the beginning of these release notes, is equally exciting. The much-anticipated re-arch is now available! The team, led by @collijk, has worked tirelessly over the past few months to put the "Auto" back in Auto-GPT, nearly doubling the code available in the master branch. The autogpt/core
folder contains the work from the re-arch project, which is now systematically making its way to the rest of the application, starting with the Configuration modules. Watch for improvements over the next few weeks. There is still much to do, so if you wish to assist, please check out this issue.
Besides the highlights above, this release cleans up longstanding Azure configuration rough edges, fixes plugin incompatibilities and plugs security. Read on for a detailed list of changes.
stable
version v0.4.3 back to master
by @lc0rp in https://github.com/Significant-Gravitas/Auto-GPT/pull/4828
docker-compose
by @kerta1n in https://github.com/Significant-Gravitas/Auto-GPT/pull/4471
ai_name
, ai_role
, and ai_goals
by @rocks6 in https://github.com/Significant-Gravitas/Auto-GPT/pull/3250
stable
into release-v0.4.4
(#4876) by @Pwuts in https://github.com/Significant-Gravitas/Auto-GPT/pull/4907
Full Changelog: https://github.com/Significant-Gravitas/Auto-GPT/compare/v0.4.3...v0.4.4
Published by lc0rp over 1 year ago
We're excited to present the 0.4.3 maintenance release of Auto-GPT! This update primarily focuses on refining the LLM command execution, extending support for OpenAI's latest models (including the powerful GPT-3 16k model), and laying the groundwork for future compatibility with OpenAI's innovative function calling feature.
As part of our ongoing commitment to refining Auto-GPT, the following commands, which we determined to be either better suited as plugins or redundant, have been retired from the core application:
As you may recall, we recently embarked on a significant re-architecting journey to future-proof the Auto-GPT project. We're thrilled to report that elements of this massive overhaul are now being integrated back into the core application. For instance, you may notice less reliance on global state being passed around via singletons.
Stay tuned for further updates and advancements in our future releases! Head over to the discussion forums or discord to share your feedback on this release, and we appreciate your continued support.
replace_in_file
command to change occurrences of text in a file by @bfalans in https://github.com/Significant-Gravitas/Auto-GPT/pull/4565
MemoryItem
s from file in JSONFileMemory
by @Pwuts in https://github.com/Significant-Gravitas/Auto-GPT/pull/4703
Full Changelog: https://github.com/Significant-Gravitas/Auto-GPT/compare/v0.4.2...v0.4.3
Published by lc0rp over 1 year ago
We're excited to present the 0.4.3 maintenance release of Auto-GPT! This update primarily focuses on refining the LLM command execution, extending support for OpenAI's latest models (including the powerful GPT-3 16k model), and laying the groundwork for future compatibility with OpenAI's innovative function calling feature.
As part of our ongoing commitment to refining Auto-GPT, the following commands, which we determined to be either better suited as plugins or redundant, have been retired from the core application:
As you may recall, we recently embarked on a significant re-architecting journey to future-proof the Auto-GPT project. We're thrilled to report that elements of this massive overhaul are now being integrated back into the core application. For instance, you may notice less reliance on global state being passed around via singletons.
Stay tuned for further updates and advancements in our future releases! Head over to the discussion forums or discord to share your feedback on this release, and we appreciate your continued support.
replace_in_file
command to change occurrences of text in a file by @bfalans in https://github.com/Significant-Gravitas/Auto-GPT/pull/4565
MemoryItem
s from file in JSONFileMemory
by @Pwuts in https://github.com/Significant-Gravitas/Auto-GPT/pull/4703
Full Changelog: https://github.com/Significant-Gravitas/Auto-GPT/compare/v0.4.2...v0.4.3-alpha
Published by merwanehamadi over 1 year ago
The 503 error has been more frequent the past hours so we added a hotfix to retry the call if this error is returned, otherwise Auto-GPT stops.
Published by lc0rp over 1 year ago
This release focuses on improving command reliability and fixing bugs from version 0.4.0. Some internal changes have been made that pave the way for the highly anticipated 0.5.0 major release.
As usual, this release cannot have happened without Auto-GPT's ever-enthusiastic community. Hop over to the discussion boards to discuss this release and more.
This time round, we're highlighting "New Contributors" before the longer changelog details, and including testers, PR-reviewers and other active community members. We may keep this new format in future releases.
replace_in_file
command to change occurrences of text in a file by @bfalans in https://github.com/Significant-Gravitas/Auto-GPT/pull/4565
master
by @Pwuts in https://github.com/Significant-Gravitas/Auto-GPT/pull/4585
split_file
from file_operations.py by @Pwuts in https://github.com/Significant-Gravitas/Auto-GPT/pull/4658
Full Changelog: https://github.com/Significant-Gravitas/Auto-GPT/compare/v0.4.0...v0.4.1
Published by Pwuts over 1 year ago
TL;DR: Users will only see a small part of the improvements made in this release. Most of the improvements were made "under the hood", where developers will see cleaner source code that is becoming increasingly easier to work with.
Highlights and notable changes since v0.3.0:
send_tweet
is REMOVEDTwitter functionality (and more) is now covered by plugins.
The Milvus, Pinecone and Weaviate memory backends were rendered incompatible by work on the memory system, and have been removed. The Redis memory store was also temporarily removed; we will merge a new implementation ASAP.
Whether built-in support for the others will be added back in the future is subject to discussion, feel free to pitch in: https://github.com/Significant-Gravitas/Auto-GPT/discussions/4280
read_file
πAuto-GPT can now read text from document files, with support added for PDF, doc, docx, CSV, HTML, TeX and more!
You can now disable set of built-in commands through the DISABLED_COMMAND_CATEGORIES
variable in .env. Specific shell commands can also be disabled using DENY_COMMANDS
, or selectively enabled using ALLOW_COMMANDS
.
We are working on re-architecting large parts of the codebase to make development easier in the long term and future-proof the project. This re-architecture is a big task and will take a while, so in the meantime, we continue merging pull requests into master
(our main development branch) and integrating contributions that help the project's overall objectives, including re-arch.
The overarching vision of the re-arch is to re-imagine Auto-GPT as a component-based framework geared towards facilitating exploration of AI agents. If the internals are not clean, progress suffers.
This vision breaks down into specific goals:
In this release, we make two major re-architecture steps:
Memory System Revamp
We have revised the vector memory storage system (#4208), laying a foundation to implement so-called "retrieval augmentation" that enhances the long-term performance of Auto-GPT.
For the time being we have removed support for all memory providers except for local (json_file
) memory. This stop-gap measure simplifies/expedites development. A Redis memory provider implementation is in the pipeline, we're waiting on the redis-om-python team to merge a fix.
Work on the memory system is tracked here: #3536
Removal of global configuration object
In this version, we've removed the global configuration object. Instead configuration is passed into components as required. Further work is ongoing to remove these 'singletons'.
stable
into master
after v0.3.1 release by @richbeales in https://github.com/Significant-Gravitas/Auto-GPT/pull/4203
gh pr checkout
by @merwanehamadi in https://github.com/Significant-Gravitas/Auto-GPT/pull/4363
Full Changelog: https://github.com/Significant-Gravitas/Auto-GPT/compare/v0.3.1...v0.4.0
Published by p-i- over 1 year ago
.env
file.plugin.post_planning
call.master
with fixes from release-v0.3
by @Pwuts in https://github.com/Significant-Gravitas/Auto-GPT/pull/3694
plugin.post_planning
by @Androbin in https://github.com/Significant-Gravitas/Auto-GPT/pull/3414
milvus_memory_test.py
mock Config
by @Androbin in https://github.com/Significant-Gravitas/Auto-GPT/pull/3424
Full Changelog: https://github.com/Significant-Gravitas/Auto-GPT/compare/v0.3.0...v0.3.1
Published by richbeales over 1 year ago
Over a week and 275 pull requests have passed since v0.2.2, and we are happy to announce the release of v0.3.0! From now on, we will be focusing on major improvements rather than bugfixes, as we feel stability has reached a reasonable level. Most remaining issues relate to limitations in prompt generation and the memory system, which will be the focus of our efforts for the next release.
Now 3rd-party developers can extend the functionality of AutoGPT by writing plugins.
Do you want Auto-GPT to access your calendar or open your garage door? Write a plugin for it!
Please note the Plugin interface is unstable and will change with potentially v0.3.1 and v0.4.0 as we add more capability
https://github.com/Significant-Gravitas/Auto-GPT-Plugins#plugins-in-the-repository
The docs have been updated and can now be found on https://docs.agpt.co/. RTFM! :)
Many proposals come in, but how to know which ones actually improve Auto-GPT? To better focus our efforts, our tests now include challenges, e.g., a memory challenge (levels 1 to 10).
AutoGPT can get up to level 4. Can any rockstar dev get it to level 5 or level 10?
Auto-GPT just became a lot easier to use with an automatic initial prompt generator.
Just say what you want, e.g., "make pong," and Auto-GPT will (attempt to) do the rest.
This feature enables the AI to engage in reflection and improve its thought, plans, and reasoning capabilities by considering its own internal evaluations. To use this feature, input the 'S' key when Auto-GPT asks for your input.
To better keep the AI on track, Auto-GPT now maintains a running summary of prior interactions.
After each interaction, this summary is updated. In this way, Auto-GPT remembers what's essential as it works.
The above are highlights out of the changes you'll see as a user, but it's just the tip of the iceberg.
Behind the scenes, the core team has been restructuring the codebase and organizing itself.
We're paying back the technical debt that was accrued during our early days. There's much more to come.
./run.sh start
-> ./run.sh
by @chenfengyanyu in https://github.com/Significant-Gravitas/Auto-GPT/pull/2523
browse_website
by @peterbanda in https://github.com/Significant-Gravitas/Auto-GPT/pull/2680
validate_json
file error when cwd != project root by @mickjagger19 in https://github.com/Significant-Gravitas/Auto-GPT/pull/2665
size
param to generate_image_with_dalle by @AndresCdo in https://github.com/Significant-Gravitas/Auto-GPT/pull/2691
download_file
command by @Nickfost in https://github.com/Significant-Gravitas/Auto-GPT/pull/2753
load_azure_config
throw if azure.yaml
does not exist by @Pwuts in https://github.com/Significant-Gravitas/Auto-GPT/pull/2758
IMAGE_PROVIDER=sd
=> IMAGE_PROVIDER=huggingface
) by @okunishinishi in https://github.com/Significant-Gravitas/Auto-GPT/pull/2802
get_ada_embedding
return type by @Androbin in https://github.com/Significant-Gravitas/Auto-GPT/pull/3263
vcr_config
scope to "session" by @Androbin in https://github.com/Significant-Gravitas/Auto-GPT/pull/3361
browse_website
headless mode for Firefox by @Irmius in https://github.com/Significant-Gravitas/Auto-GPT/pull/2816
clone_repository
to conform to URL validation by @SBNovaScript in https://github.com/Significant-Gravitas/Auto-GPT/pull/3150
list_files
api signature change in data_ingestion.py
and lo⦠by @valayDave in https://github.com/Significant-Gravitas/Auto-GPT/pull/3601
read_file
by @sidewaysthought in https://github.com/Significant-Gravitas/Auto-GPT/pull/3173
validate_json
scheme path by @k-boikov in https://github.com/Significant-Gravitas/Auto-GPT/pull/3631
stable
-> master
by @Pwuts in https://github.com/Significant-Gravitas/Auto-GPT/pull/3677
Full Changelog: https://github.com/Significant-Gravitas/Auto-GPT/compare/v0.2.2...v0.3.0
Published by richbeales over 1 year ago
Welcome to Auto-GPT release 0.2.2. This is largely a release focussed on stability over 0.2.1. More exciting stuff will becoming in 0.3.0!
evaluate_code
to analyze_code
by @willcallender in https://github.com/Significant-Gravitas/Auto-GPT/pull/1371
Full Changelog: https://github.com/Significant-Gravitas/Auto-GPT/compare/v0.2.1...v0.2.2
Published by richbeales over 1 year ago
create_chat_message
and generate_context
methods of the chat module by @OmriGM in https://github.com/Significant-Gravitas/Auto-GPT/pull/1700
Full Changelog: https://github.com/Significant-Gravitas/Auto-GPT/compare/v0.2.0...v0.2.1
Published by richbeales over 1 year ago
We have changed the way that Auto-GPT is launched, it is now wrapped in a module which will allow us better testing and make
the code more maintainable and extensible in the future.
NOTE: Auto-GPT (in this release) currently requires Chrome to be installed in order to browse the web.
Other dependencies have also been introduced - please re-run pip install -r requirements.txt
Full Changelog: https://github.com/Torantulino/Auto-GPT/compare/v0.1.3...v0.2.0
Published by richbeales over 1 year ago
Daily release, encompasses the merging of PRs in "Batch 4"
https://github.com/Torantulino/Auto-GPT/issues/1015
Published by richbeales over 1 year ago
Daily release, based on the merge of PRs in Batch 3
https://github.com/Torantulino/Auto-GPT/issues/709
Published by p-i- over 1 year ago
Many PRs:
Remainder of Batch2 - https://github.com/Torantulino/Auto-GPT/issues/673
Some of Batch 3 - https://github.com/Torantulino/Auto-GPT/issues/709