Engineering
Software engineering team
An issue comes in, a reviewed PR goes out. The same path runs every time.
Issue assigned
implement
open PR
human PR review· required
merged
GitHubLinear
bobiOPEN SOURCE · MIT
The agent harness for enterprise work.
A team of agents on your events, running your work in real time, with guardrails written into the workflow as nodes.
Open source · MIT · self-hosted · built on the Claude Agent SDK
WHAT IT IS
A realtime agent team running your workflows. Not an agent waiting for a chat.
Fig. 01 · Event-driven agent flowLive
◇ ticket lands▢ manager + agents▤ yaml workflow◆ human approval✓ shipped
Moda Labs · bobi
Responds in real time
EVENT BUS
GitHub, Linear, Slack, email, and more all through one event bus. A job can start the moment an event happens, through a chat, or via cron.
Configurable agent teams
MANAGER + AGENTS
A manager plus the agents a job needs, pointed at any workflow. Not one assistant. A team with roles.
Deterministic guardrails
WORKFLOW GATE
Your process is a YAML workflow the agents follow. Review, approval, and spend limits are nodes in the workflow, not suggestions in a prompt.
WHY IT’S DIFFERENT
What makes bobi different.
Most harnesses give a model a set of tools. bobi gives it a team, a process, and guardrails that hold.
01
Responds in real time
GitHub, Linear, Slack, email, and more land on one event bus, normalized to a single stream. Start a run from an event, a chat, or a cron, all in real time.
event-bus.yaml1 stream
sources: # all on one bus github issue · pr · review linear ticket · comment slack alert · mention email inbound chat message · command cron schedule webhook * # and anything else↳ any trigger starts the run in ~0.2s02
Configurable agent teams
Name a manager and the agents the job needs, each with its own tools and role. Point the team at any workflow and change the roster in one file.
team.yamlRoles
manager: orchestratoragents: engineer: { tools: [repo, ci] } reviewer: { tools: [repo] } qa: { tools: [ci, browser] }workflow: sdlc.yaml03
Deterministic guardrails
Review steps, approvals, and spend limits are nodes the run has to pass. The model cannot talk its way around them, because they live in the graph, not the prompt.
guardrails.yamlEnforced
workflow: - run: implement - gate: human_review # required approvers: [maintainer] - limit: spend # node, not a hint max_usd: 50 on_exceed: halt - done: mergedPlus three more reasons it holds up
04
Runs on your Claude plan
Every agent runs Claude Code, so the whole system runs on your Claude Pro or Max plan. No per-token API bills, so you can orchestrate a full team economically.
05
Defined in YAML
Your process is a versioned YAML workflow, so every run follows the same path. Read, review, and diff it like code.
06
Teams that work together
Agent teams hand off to each other, so your engineering team and your marketing team collaborate directly.
USE CASES
Need realtime agents with a dependable process? bobi has your back.
FAQ
Common questions.
What is bobi?
bobi is an open-source agent harness: a configurable team of autonomous agents that responds to your events in real time, with guardrails written into the workflow as nodes. It is MIT-licensed, self-hosted, and built on the Claude Agent SDK.
How is bobi different from a chatbot or a single agent?
bobi still answers a chat, but it also wakes on real events (tickets, emails, alerts, PRs) and on a schedule, all through one event bus. And instead of one assistant, it runs a configured team of agents through a YAML workflow where review, approval, and spend limits are nodes, not suggestions in a prompt.
How does bobi handle memory, context, and self-improvement?
bobi's memory works on two levels. A global knowledge base builds semantic context from what it has seen across your enterprise tools, so any agent team can pull relevant background instead of starting cold. On top of that, learning memory lets it remember and improve from the outcomes and feedback of past runs, getting sharper at your workflows over time. We are actively deepening both.
How does bobi compare to Hermes?
Hermes is Nous Research's self-improving, open-source agent: a single autonomous agent that lives on your server, carries persistent memory, writes and refines its own skills, and gets more capable the longer it runs. bobi shares that learning memory and adds a global knowledge base across your enterprise tools, but it is built around a team and a process: you point it at a specific workflow, it triggers on your events, runs a manager plus the agents a job needs, and enforces review, approval, and spend limits as workflow nodes. Think one self-improving agent versus a configured team running your process with guardrails. The two are complementary.
How does bobi compare to OpenClaw?
OpenClaw is a self-hosted, MIT-licensed gateway that brings an AI assistant into your chat apps (Slack, Telegram, WhatsApp, iMessage, Discord, and more) with tool use, sessions, memory, and multi-agent routing. You message it and it acts. bobi shares that open-source, self-hosted, multi-agent foundation, but it is event-first rather than chat-first: it runs your enterprise queues on their own from tickets, alerts, and PRs (plus chat and cron) through a defined YAML workflow, with review, approval, and spend limits as guardrail nodes. Reach for OpenClaw to put an assistant in your messaging; reach for bobi to run a workflow with a process and a human gate where it matters.
Is bobi open source?
Yes. bobi is MIT-licensed and runs entirely on your own infrastructure.
What is bobi built on?
bobi is built on the Claude Agent SDK.
What can bobi automate?
Any queue someone watches all day following a process: software engineering (issue in, reviewed PR out), inbound support, data pipeline ops, recruiting and screening, inbound sales, accounts payable, and market intelligence. Each runs with its own trigger, steps, and human gate.
