SuperAGI

Toolkits let agents take authenticated action in outside systems. Search runs through Google Search, DuckDuckGo, Searx and SerpAPI, and developer toolkits cover GitHub, including Fetch PR and Review PR, Jira and a Coding toolkit.

Productivity toolkits cover Notion, Google Calendar and a File Manager, Email sends messages with attachments, Twitter and Instagram cover social posting, DALL-E and Stable Diffusion generate images, Apollo.io supplies sales data and Google Analytics supplies analytics. Credentials are set per toolkit in the GUI, with OAuth where the service requires it.

Developers can add custom toolkits through a linked GitHub repository, a cookie-cutter template and a community tools repository, so the catalog is open to additions.

Sourcegithub.com/TransformerOptimus/SuperAGI README and release notes v0.0.9 through v0.0.12read 2026-09-01

Agents run a ReAct reasoning-and-acting loop, and the runtime handles the control flow rather than leaving it to the developer. Predefined agent workflows can be chosen at provisioning as sequential instruction sets, and Sales Engagement and Recruitment ship as worked multi-step examples. A wait block lets a workflow pause for a set time before continuing.

A task queue sets priorities, a configurable max-iteration limit caps each run, looping detection heuristics flag an agent that is stuck, and provider errors such as rate limits, authentication failures and exceeded quotas are shown to the operator instead of failing silently. Several agents can run at the same time.

Sourcegithub.com/TransformerOptimus/SuperAGI release notes v0.0.10 and v0.0.13, and README architecture diagramsread 2026-09-01

Agents can work from the customer's own content, retrieved at run time. Files in pdf, docx, pptx, csv, txt and epub are uploaded through the resource manager, turned into vector embeddings by the LlamaIndex integration and queried by the agent through the Resource Query tool.

Knowledge Embeddings can also be created from the customer's own sources or installed from the marketplace, hosted in a Pinecone, Qdrant, Weaviate or Redis index the customer connects by URL, and searched semantically at run time through the KnowledgeSearch toolkit. A Knowledge Console logs embedding performance, queries and token consumption.

Sourcegithub.com/TransformerOptimus/SuperAGI release notes v0.0.7, v0.0.9 and v0.0.11, superagi.com knowledge-embeddings articleread 2026-09-01

Restricted mode is chosen per agent as a permission type at provisioning, and an agent running under it must get human approval before critical actions such as writing files, sending emails and moving to the next iteration. Requests appear in the Action Console, where the operator can Approve, Give feedback or Deny, so the customer decides which kinds of action need sign-off, who gives it and what happens while the request waits. A configurable max-iteration limit caps a run separately from approvals.

Sourcegithub.com/TransformerOptimus/SuperAGI release notes v0.0.4, v0.0.6 and v0.0.10read 2026-09-01

Authentication runs through GitHub OAuth, which also protects the APIs, alongside a configurable auth secret and an option to restrict list creation, and the project runs CodeQL code scanning in its own pipeline.

The framework has no access model for who inside an organization can make an agent do what, and it publishes no compliance page or certification of its own. Its logs record telemetry, and SuperAGI names no audit trail. The GDPR, SOC 2 and HIPAA claims made for the company's separate commercial platform belong to that product, not to the framework. Self-hosting means the operator supplies the surrounding controls.

Sourcegithub.com/TransformerOptimus/SuperAGI release notes v0.0.4 and v0.0.7, config_template.yamlread 2026-09-01

The Agent Performance Monitoring dashboard reports organization metrics for agents, runs and tokens, per-model breakdowns of runs and consumption, and agent-level analytics covering total runs, average tokens per run, total calls and run time.

Model, Tool and Knowledge Consoles each keep their own logs, so a finished run can be traced step by step through which model was called, which tools ran and which embeddings were queried.

A real-time activity feed shows execution as it happens, APM keeps data for deleted agents and runs, and the resource manager keeps a per-run output directory that downloads as a zip.

Sourcegithub.com/TransformerOptimus/SuperAGI release notes v0.0.8, v0.0.9 and v0.0.12read 2026-09-01

Short-term memory is a rolling window sized to the model's token limit. Context that falls outside it is condensed into a Long Term Summary, and the two are combined into an Agent Summary that goes into each reasoning step, weighted seventy-five to twenty-five. Long-term memory persists in a vector database, with Qdrant available as an LTM backend alongside Pinecone, Weaviate and Redis. Tool Memory stores every tool response from a run as vectorized chunks in Redis for semantic retrieval, though only the Thinking Tool can read it.

Sourcegithub.com/TransformerOptimus/SuperAGI release notes v0.0.8, v0.0.10 and v0.0.11read 2026-09-01

The framework is MIT-licensed and self-hosted with Docker Compose on the customer's own machine or infrastructure, and a separate GPU compose file supports local large language models and multi-GPU execution, so neither the platform nor the model has to leave the environment. A one-click Digital Ocean deployment runs the image on a self-hosted cloud machine and updates itself with each release. Storage can be local disk or S3, and the database and vector store belong to the operator. A hosted SuperAGI Cloud is also available for teams that prefer it.

Sourcegithub.com/TransformerOptimus/SuperAGI README installation section and release notes v0.0.4, v0.0.12 and v0.0.14read 2026-09-01

A browsable marketplace carries toolkits, agent templates, knowledge embeddings and models, reachable from the product and at its own subdomain. SuperCoder builds end-to-end coding projects from goals and instructions.

Predefined agent workflows include Sales Engagement, which fetches prospects from Apollo and drafts personalized outreach for each row, and Recruitment, which compares uploaded CVs against a job description and sends shortlist or rejection mail.

Customers can save templates from any run and publish them to the marketplace after a validated run, so the library grows from the community as well as from SuperAGI.

Sourcegithub.com/TransformerOptimus/SuperAGI release notes v0.0.6, v0.0.7, v0.0.10 and v0.0.12, marketplace.superagi.comread 2026-09-01

An agent can be scheduled for a single future run or for a recurring run at a set interval, with optional expiry by deadline or by total run count, and schedules can be stopped, edited and saved, so the agent starts without anyone typing.

Webhooks fire outward when an agent is running, paused, completed, terminated or reaches max iterations, which notifies external systems but does not start work. People reach the agent only through the framework's own web interface at localhost or the hosted console, and there is no inbound channel from email, chat or a team tool. Email, Twitter and Instagram are outbound toolkits the agent acts through, not channels that reach it.

Sourcegithub.com/TransformerOptimus/SuperAGI release notes v0.0.8 and v0.0.12read 2026-09-01

Providers are configured in account settings, and a model is chosen per agent from a dropdown at provisioning. GPT-4-32k, gpt-3.5-turbo-16k and Google PaLM 2 Bison are supported, and a model factory class is built to admit more. HuggingFace and Replicate integrations let the customer bring any public or self-hosted model by entering the model name and inference endpoint, a models marketplace offers preconfigured entries, and local large language models run on the customer's own weights with multi-GPU support.

Sourcegithub.com/TransformerOptimus/SuperAGI release notes v0.0.7, v0.0.8, v0.0.11 and v0.0.14read 2026-09-01

Public APIs, reachable from cloud or local base URLs, let developers create, pause, resume and update agents, start runs and retrieve run details and input or output files. API keys are generated in settings, and the full reference is published as a Postman collection. Python and NodeJS SDK client libraries install with pip or npm and have their own documentation section.

Webhooks send HTTP callbacks to a user's endpoint on five events, when an agent is running, paused, completed, terminated or reaches max iterations. Custom tools can be added by linking a GitHub repository from the frontend, with a cookie-cutter template and a community tools repository.

Sourcegithub.com/TransformerOptimus/SuperAGI release notes v0.0.11, v0.0.12 and v0.0.13, and documenter.getpostman.com SuperAGI collectionread 2026-09-01

The Coding toolkit ships a Write Unit Tests tool, and the SuperCoder sequence runs WriteSpec, WriteCode, ImproveCode and then WriteTest, so tests are written against the customer's generated project. SuperAGI does not say whether those tests are run, or whether a failing test blocks delivery or is reported back. Agent Performance Monitoring, the Model, Tool and Knowledge consoles and the activity feed report what happened after a run. There is no evaluation harness, scored comparison between agent versions or preview run against a held-out set.

Sourcegithub.com/TransformerOptimus/SuperAGI release notes v0.0.6 and v0.0.9read 2026-09-01

A Web Scraping tool in the toolkit set pulls data from webpages, alongside a Web Interaction toolkit. Both read the web through scraping and wired-in access and do not drive a real interface. SuperAGI has no hosted or local browser, desktop session or remote computer control, and its search tools are API-based.

Sourcegithub.com/TransformerOptimus/SuperAGI release notes v0.0.4 and README toolkit listread 2026-09-01