Your team, at full strength.

Chief is an open-source AI chief of staff that your team shares in Slack. People brief and correct it in channels or DMs; Chief keeps the context, turns each request into tracked work with a clear owner and reports back where the request started. A launch thread can become a sourced brief with owners and open decisions, without someone rebuilding the history each time. Chief can send separate jobs to AI workers it supervises and keep following up until the team reviews or closes the work. It asks before an external send, a production change or another hard-to-reverse action.

Chief keeps work moving across messages, threads and sessions; a chatbot normally answers the prompt in front of it. A teammate can write “make the tests smaller” to redirect a running investigation without starting it again. Chief follows up on work that is still open, and the original thread remains the place where the result lands. The team keeps control of access, spending limits and actions that need approval.

Everyone you approve can steer the same Chief in the Slack threads it can access. An engineer can add a technical constraint to a running job without asking a teammate to restate the original request. Chief tracks who asked for what, keeps the plan and results visible, and accepts corrections from the people in the conversation. If two directions conflict, Chief shows the conflict and asks who has the decision instead of choosing silently. Each worker gets one clearly scoped job, while the main Chief agent keeps the decisions and results visible in the shared conversation.

Yes. Leaders can set Chief’s writing tone and review criteria, and tell it when to follow up. Chief can flag a product decision hidden in a pull request and explain it for the responsible lead to decide. It applies the leader’s instructions when it reviews a draft or reminds an owner. Chief drafts external messages and waits for approval when the final judgment or send belongs to a person. Corrections are saved to Chief’s running record, so the next review starts from the latest direction.

Chief’s workers can run on Codex, Claude Code or Pi, using the agents, skills, connectors and tools your company approves. One worker can research while another edits code, and Chief keeps their progress in the original Slack thread. The main Chief agent runs on Claude Code with your Claude subscription or on Hermes with any model provider. Chief gives each worker the relevant context and a checklist of what it must verify, then brings the result back to the main agent for review. You decide which engines and connections are available; a worker does not gain extra access just because its engine supports it.

Chief lives in Slack and runs on macOS or Linux. A small team can keep Chief on an always-on Mac or Linux host and use it from the Slack rooms where the work already happens. It is open source, and its memory and list of in-progress tasks stay on the host you choose. The main agent can use Claude Code on your existing Claude subscription or Hermes with any model provider, while its supervised workers can run on Codex, Claude Code or Pi. You control the Slack app, model credentials, connected tools and the rooms Chief can access.

Run the main Chief agent with Claude Code on a Claude subscription or with Hermes and any model provider. The main agent can stay on Claude while a code review goes to Codex and a short summary goes to Pi. For worker jobs, Chief supports Codex, Claude Code and Pi; the provider and model depend on the worker configuration you set. You choose the profiles, budgets and tools that are available, and Chief uses only those options. Chief can watch model spend and ask before it exceeds a limit you set.

Chief works inside the access and authority you give it. You choose the Slack rooms, tools, accounts and credentials it can use, and each worker receives only the context and access its job needs. Chief can prepare a production change and its test evidence, but an authorized person still decides whether that change ships. It stops for approval before customer messages, production changes, costly work or other hard-to-reverse actions. Chief reports missing access, uncertain claims and failed checks instead of filling the gaps with guesses.

Chief needs a Slack app, a macOS or Linux host and only the rooms and tools required for the work you give it. A code-review worker can receive one repository and the pull request in question without receiving access to customer systems. You choose the approved Slack channels and DMs, connect the relevant repositories, documents or services, and keep credentials in their supported stores. Chief passes a worker the smallest useful slice of context and access for its job, then reviews what comes back. Missing access is reported plainly and does not become permission to use another account or destination.

Chief carries the work context you give it across threads and sessions. A weekly plan can cite the meeting transcript, the merged pull requests and the commitments that are still open. That context can include Slack channels and DMs, meetings, code, tickets, documents, Drive files, worker results, decisions, owners and promises. Chief’s memory and task list stay on the host you choose, so a later job can start from the last verified state. Access still follows the rooms, people and tools you approved, and Chief marks what it could not verify.