Researching a subject or preparing a report involves more than a single question. It means finding information, working through it and producing something useful. We built Agenhood to bring those steps together. It is an open-source platform for running AI agents: software assistants that use tools to research, write and process data, with a place to keep their work and continue it later.
Visit agenhood.devThe starting point is a job someone needs done: compare suppliers, summarize a collection of articles or organize a data file. In Agenhood, an agent combines an AI model with instructions and tools for that job. Those tools let it find information, work with files and produce a result. Different assignments can have different instructions and capabilities.
Carrying out an assignment requires somewhere to collect material and save the output. Each Agenhood agent has its own workspace for that purpose. For a research brief, it can gather sources, save notes and turn the findings into a document. The assignment produces files that can be reviewed, reused or developed further.
That workspace remains available after a task ends. Agenhood keeps the agent's files, instructions and task history across sessions, so a follow-up assignment can use what already exists. A draft can become a finished report. New findings can be compared with earlier research. Continuity comes from keeping the material the next task will need.
Some assignments need to happen again. A weekly digest, for example, needs fresh research and the previous edition for comparison. Agenhood can run a saved assignment on a schedule, using the instructions and material already in place. This extends a useful one-off task into recurring work without someone having to start every run.
A larger assignment may benefit from several agents with different responsibilities. One can collect sources, another analyze them and a third prepare the report. Agenhood connects these stages into a workflow: an ordered sequence of tasks that passes files from one stage to the next. The complete workflow can also run on a schedule.
Agenhood is self-hosted, meaning it runs on a server controlled by the person or organization using it. They choose the model providers and decide how agents are configured. Task history records what happened, limits control resource use, and saved versions make it possible to recover earlier files.
Docker provides an isolated place for each agent to run tools and produce files. Workspaces persist across pauses and restarts, allowing ongoing assignments to retain their material without keeping every agent active all the time.
A Python and FastAPI backend manages recurring assignments, workflow steps and their results. Defined handling for completion, failure and cancellation keeps unfinished steps from being treated as finished work.
A shared runtime connects different agent engines to the same task system. The web application and external software use a common API, so direct assignments and automated requests share the same controls and working history.
A report, a comparison and a data extraction have different definitions of done. We built Agenhood around configurable instructions, tools and outputs so the assignment can set that definition. Its breadth comes from those building blocks, with room for both everyday tasks and more specialized work.
A saved source, a research note or an earlier draft can matter more to the next assignment than a fresh start. Persistence makes those materials available over time. That shaped Agenhood's approach to workspaces, recurring tasks and recovery: the product needs to support the life of an assignment across many runs.
Long-running work needs clear limits, a record of what happened and a way to recover files. We built those controls alongside scheduling and agent coordination, and tested interrupted runs as well as successful ones. They make it possible to hand over an assignment while retaining responsibility for the result.