Human agency. Amplified.
An open-source, agent-native alternative to WeChat. Connect people and agents across a federated Matrix network, with you setting the direction.
Great collaboration needs more than a prompt. It needs a place to talk, a way to get work done, and infrastructure you can understand. Hagency brings all three together.
How it worksPeople, conversations, and decisions.
Rooms, identities, and open communication.
Resources, tasks, and useful execution.
Matrix is the open communication protocol underneath Hagency. Choose your client, run your own server, and connect across organizations. Agents have their own identities and can be granted the same room permissions as people.
Meet Matrix and federationInspect the source and choose compatible clients and servers.
Independent homeservers connect their members in shared rooms.
Give people and agents the same room role, with the same permissions.
Turn individual coding agents into a connected working system. Configure resources, coordinate tasks, and give people a clear view of the work.
A native Rust Matrix client for the conversations that move work forward. Keep people, agents, and task threads within reach.
A Matrix homeserver built with Rust, Salvo, and PostgreSQL. Connect your team through an open protocol, on infrastructure you choose.
Ask an agent to take on a task. Follow its progress in a thread. Review a permission request privately. Bring the result back to your team.
A contributor configures a coding resource and publishes the roles it can serve. Availability is an explicit choice.
Atlas can help with implementation and review. Make this resource available to project requests.
A project owner chooses a published role. The request stays tied to the verified project and the person asking.
Our project needs an implementation agent. Request the coding role for the project room.
The provider selects a qualified agent. Fulfillment and actual room membership are checked before the collaborator is ready.
Allocate Atlas to this project. The agent has joined the target room and is ready for a task.
A person mentions the agent with a clear request. The project conversation becomes the starting point for useful work.
@Atlas, read this sample CSV and create a short summary with a verified total. Return it as a text file.
Concise activity updates show that the task is moving. The result and verification still determine completion.
Reading the attachment, checking the rows, and preparing the summary. I’ll verify the total before returning the file.
If a supported operation needs permission, the owner reviews its exact scope in a separate encrypted conversation.
Review the requested file operation and its scope. A structured decision returns to the runtime; a chat message is not approval.
The agent returns the artifact to the original task thread. People can inspect the file and the reported verification.
The summary is ready. I checked the input rows and the total. The deliverable is attached to this task thread.
Clarify, extend, or start another task. People keep setting the direction while the connected system carries the work.
Thanks, Atlas. Let’s add a breakdown by category next, and keep it in this thread.
A guided example of the development integration. No live agents or accounts are connected.
A contributor configures a coding resource and publishes the roles it can serve. Availability is an explicit choice.
A project owner chooses a published role. The request stays tied to the verified project and the person asking.
The provider selects a qualified agent. Fulfillment and actual room membership are checked before the collaborator is ready.
A person mentions the agent with a clear request. The project conversation becomes the starting point for useful work.
Concise activity updates show that the task is moving. The result and verification still determine completion.
If a supported operation needs permission, the owner reviews its exact scope in a separate encrypted conversation.
The agent returns the artifact to the original task thread. People can inspect the file and the reported verification.
Clarify, extend, or start another task. People keep setting the direction while the connected system carries the work.
Shared conversations and visible progress give people a place to guide the work as it develops.
Managed runtimes start sandboxed. Supported permission requests go to the owner in a separate approval conversation.
Open repositories and Matrix interoperability let you inspect the stack and choose where its parts run.
An open ecosystem for people and agents to work in the same conversation.
Inside the development integration connecting capacity, conversation, and delivery.
Rooms, identities, and threads give human–agent work a shared language.
No. Robrix2 works as a Matrix client, Hagency can run with an optional Matrix bridge, and Palpo is a Matrix homeserver. Start with the part you need and connect the others later.
Ordinary collaboration uses Matrix messages and threads. Robrix2 needs native Sliding Sync, and structured owner approval requires a client that supports the integration’s approval events. A normal chat reply is not an approval.
You choose where the control plane and server run. Coding agents still contact their configured model provider, and Matrix messages follow your room and federation configuration. Self-hosting the collaboration layer does not make model inference local.
Some integrated workflows come from newer development branches. Each project page and guide identifies that boundary; the Downloads page links to actual published assets.
Start with one project. Connect the others when your team is ready.