agentbus

agentbus

agentbus is a small NATS JetStream based task bus for distributed agent programs.

It is designed for this architecture:

                              +-------------------+
                              |       human       |
                              +---------+---------+
                                        |
                                        v
                              +-------------------+
                              |    main agent     |
                              +----+---------+----+
                                   |         ^
                    publish tasks  |         |  read results
                                   v         |
+----------------------------------+---------+----------------------------------+
|                         public NATS JetStream bus                             |
|                                                                               |
|        +---------------------+                 +----------------------+       |
|        |    tasks stream     |                 |    results stream    |       |
|        +----------+----------+                 +----------+-----------+       |
|                   |                                       ^                   |
+-------------------|---------------------------------------|-------------------+
                    |                                       |
      deliver tasks |                                       |  publish results
                    v                                       |
+-------------------+---------------------------------------+-------------------+
|                              agentbus workers                                 |
|                                                                               |
|        +------------------------+             +------------------------+      |
|        |      coder worker      |             |    reviewer worker     |      |
|        |  invokes coder agent   |             | invokes reviewer agent |      |
|        +------------------------+             +------------------------+      |
+-------------------------------------------------------------------------------+

Routing summary:

• The main agent publishes tasks for target agents and may read returned results.
• JetStream stores and delivers tasks to matching agentbus workers.
• Workers invoke configured agents and publish results.

Design goals

• No bot-to-bot chat dependency.
• No direct inbound access needed for worker machines.
• Generic agent command integration through TOML chat_cmd.
• Worker and publisher configuration via TOML files.
• NATS subjects keep routing explicit and permissionable.
• Durable task delivery through JetStream, not plain fire-and-forget pub/sub.

Requirements

Server side:

• nats-server with JetStream enabled.
• nats CLI for stream setup and debugging.
• A reachable TCP port for NATS clients. The examples use non-default 7678 instead of NATS default 4222.

Worker side:

• Python >= 3.11.
• Network access from the worker machine to the NATS server.
• A one-shot agent command configured as TOML chat_cmd.

Layout

.
├── agentbus/
│   ├── __init__.py      package metadata
│   ├── cli.py           agentbus command-line entrypoint
│   ├── config.py        TOML configuration
│   ├── messages.py      task/result schema and prompt builder
│   ├── publish.py       task publishing helpers
│   ├── result.py        result reading helpers
│   └── worker.py        NATS JetStream worker runtime
├── config/
│   ├── agentbus.worker.example.toml
│   └── nats-server.conf
├── deploy/
│   ├── launchd/com.agentbus.worker.plist
│   ├── supervisor/agentbus-worker.conf
│   └── systemd/agentbus-worker.service
├── scripts/
│   └── stream-setup.sh
├── skills/
│   └── agentbus/SKILL.md
├── tests/
│   └── test_*.py
├── LICENSE
├── README.md
├── pyproject.toml
├── requirements-dev.txt
└── requirements.txt

1. Configure the NATS server

• NATS server source and releases
• NATS CLI source and releases
• Official NATS installation guide

Simple install examples:

# macOS
brew install nats-server nats-io/nats-tools/nats

# Go toolchain
go install github.com/nats-io/nats-server/v2@latest
go install github.com/nats-io/natscli/nats@latest

# Linux packages / Docker / binaries
# See the official installation guide above for the current commands.

Verify both commands are available:

nats-server --version
nats --version

Copy the sample config to the server's NATS config path. The upstream systemd example uses /etc/nats-server.conf; some distro packages may use a different path, so match the service you install.

sudo mkdir -p /data/nats /etc/nats/tls
sudo cp config/nats-server.conf /etc/nats-server.conf

Edit the config before starting the server:

sudo $EDITOR /etc/nats-server.conf

At minimum, change these values:

main password
coder password
reviewer password
client port, if `7678` is not appropriate
TLS cert/key paths, if public internet clients will connect
jetstream.store_dir, if /data/nats is not appropriate

Start the server with that config:

nats-server -c /etc/nats-server.conf

If you use a project-specific filename such as /etc/nats/agentbus.conf, start with that exact path instead:

nats-server -c /etc/nats/agentbus.conf

For service-managed deployments, make sure the service runs the same -c path and that its OS user can write to jetstream.store_dir.

The sample config defines three users:

main   publishes tasks and subscribes to central results
coder   subscribes to agentbus.coder.tasks and publishes results
reviewer    subscribes to agentbus.reviewer.tasks and publishes results

It also enables JetStream:

jetstream {
  store_dir: "/data/nats"
  max_mem_store: 256MiB
  max_file_store: 10GiB
}

accounts {
  AGENTBUS: {
    jetstream: enabled
    users: [ ... ]
  }
}

The top-level jetstream block turns on JetStream for the server. Because this sample uses named accounts, the AGENTBUS account must also have jetstream: enabled; otherwise stream creation fails with JetStream not enabled for account (10039).

NATS stores JetStream data under a jetstream/ child directory of store_dir, so this example writes data under /data/nats/jetstream. Avoid setting store_dir to a path that already ends in jetstream, or you will get a nested jetstream/jetstream directory.

TLS

For public internet deployments, avoid exposing non-TLS nats:// connections. Use TLS or put the NATS client port behind a VPN/private network.

To enable TLS, add certificate and key paths to the NATS config:

tls {
  cert_file: "/etc/nats/tls/fullchain.pem"
  key_file: "/etc/nats/tls/privkey.pem"
}

Then switch client URLs from nats:// to tls://:

nats://main:main_password@agentbus.example.com:7678
tls://main:main_password@agentbus.example.com:7678

Important network notes:

• Expose the NATS client port, 7678 in these examples, only to machines that need to connect. The NATS default is 4222; using a non-default port reduces scanner noise but is not a security boundary.
• Keep the monitoring port 8222 private or bind it only to localhost/VPN.

2. Create JetStream streams

After the NATS server is running, create the task and result streams.

Use a user with JetStream API permission. In the sample config, main has $JS.API.> access. The example below assumes you run this on the NATS server host:

./scripts/stream-setup.sh 'nats://main:main_password@127.0.0.1:7678'

This creates:

AGENTBUS_TASKS     subjects: agentbus.*.tasks    max age: 7d
AGENTBUS_RESULTS   subjects: agentbus.*.results  max age: 30d

You can inspect the streams with:

nats --server 'nats://main:main_password@127.0.0.1:7678' stream ls
nats --server 'nats://main:main_password@127.0.0.1:7678' stream info AGENTBUS_TASKS
nats --server 'nats://main:main_password@127.0.0.1:7678' stream info AGENTBUS_RESULTS

3. Install the worker

From PyPI:

python3 -m venv env
source env/bin/activate
pip install --upgrade pip
pip install agentbus

From a source checkout:

python3 -m venv env
source env/bin/activate
pip install --upgrade pip
pip install -e .

Requires Python >= 3.11. If python3 points to an older interpreter on your machine, use a versioned command such as python3.11 or python3.14.

For development and tests:

pip install -e ".[dev]"
python -m pytest tests -q

4. Configure the worker

Prefer a config file:

mkdir -p ~/.agentbus
cp config/agentbus.worker.example.toml ~/.agentbus/config.toml
$EDITOR ~/.agentbus/config.toml
chmod 600 ~/.agentbus/config.toml

If --config is omitted, agentbus worker run checks:

./agentbus.toml
~/.agentbus/config.toml
/etc/agentbus/agentbus.toml

Required worker fields:

[agent]
id = "coder"
# {input} is required and marks where AgentBus inserts the generated prompt.
chat_cmd = ["agent-cli", "chat", "--oneshot", "{input}"]

[worker]
# Use tls:// instead if the server TLS block is enabled.
server_url = "nats://coder:coder_password@agentbus.example.com:7678"

A fuller example:

[agent]
id = "coder"
chat_cmd = ["agent-cli", "chat", "--oneshot", "{input}"]
extra_instruction = ""

[worker]
# Use tls:// instead if the server TLS block is enabled.
server_url = "nats://coder:coder_password@agentbus.example.com:7678"
task_timeout_seconds = 1800
max_task_bytes = 1048576
reconnect_time_wait_seconds = 2
max_reconnect_attempts = -1

[log]
dir = "~/.agentbus/logs"
level = "INFO"
max_bytes = 104857600
backup_count = 5

Worker routing fields and JetStream stream names are derived or fixed, not configured. This keeps worker config aligned with agentbus task publish, which also derives task and result subjects from agent ids. Multiple workers with the same agent.id are replicas of the same logical agent and share the same durable consumer/progress.

chat_cmd must be a TOML array of strings. AgentBus rejects string-form commands so the generated multi-line prompt is always inserted as one explicit argv argument, never shell-parsed.

# Prompt between flags.
chat_cmd = ["agent-cli", "run", "--prompt", "{input}", "--json"]

# Hermes example. Keep `-Q` before `-q`, or put it after `{input}`;
# `-q` must be followed immediately by the query text.
chat_cmd = ["hermes", "chat", "-Q", "-q", "{input}"]

AgentBus intentionally does not use environment variables for worker configuration. Put worker settings in TOML and pass --config when you do not want the default path.

5. Run the worker

Foreground mode:

agentbus worker run --config ~/.agentbus/config.toml

For long-running deployment, use one of the included templates:

deploy/systemd/agentbus-worker.service
deploy/launchd/com.agentbus.worker.plist
deploy/supervisor/agentbus-worker.conf

Before installing a service, edit the template paths, user, working directory, and config path for the target machine.

6. Publish a task

Read the latest result in one terminal:

agentbus result get \
  --server-url 'nats://main:main_password@agentbus.example.com:7678' \
  --agent main

To keep watching after reading recent history, add --watch. --limit has the same meaning whether or not --watch is set: read the latest N stored results first. See CLI reference for argument details.

agentbus result get \
  --server-url 'nats://main:main_password@agentbus.example.com:7678' \
  --agent main \
  --limit 20 \
  --watch

Publish a task in another terminal:

agentbus task publish \
  --server-url 'nats://main:main_password@agentbus.example.com:7678' \
  --to coder \
  --to reviewer \
  --from main \
  --reply-to main \
  'hello'

Publishing uses CLI arguments instead of a TOML file because it is a short one-shot command. See CLI reference for argument details.

Examples:

agentbus task publish \
  --server-url 'nats://main:main_password@agentbus.example.com:7678' \
  --to coder \
  'hello'

agentbus task publish \
  --server-url 'nats://main:main_password@agentbus.example.com:7678' \
  --to coder \
  --task-type batch \
  '[{"url":"https://example.com"}]'

The --to coder and --to reviewer options map directly to these task subjects:

agentbus.coder.tasks
agentbus.reviewer.tasks

If the target workers are running, agentbus result get --agent main should receive task.result messages from:

agentbus.main.results

CLI reference

agentbus task publish

agentbus result get

Message reference

Subjects

Recommended convention:

Agent IDs are used literally in subjects.

agentbus.<agent_id>.tasks       tasks for one worker agent
agentbus.<agent_id>.results     optional direct result subject per agent
agentbus.main.results           central result subject for the coordinator
agentbus.<agent_id>.heartbeat   optional health events

Examples:

agentbus.coder.tasks
agentbus.reviewer.tasks
agentbus.main.results

Task message

{
  "id": "task-20260518-0001",
  "from": "main",
  "to": "coder",
  "reply_to": "main",
  "type": "task.request",
  "task_type": "review_pr",
  "payload": {
    "content": "Review PR org/repo#123"
  }
}

Result message

{
  "id": "result-uuid",
  "type": "task.result",
  "status": "completed",
  "task": {
    "id": "task-20260518-0001",
    "from": "main",
    "to": "coder",
    "reply_to": "main",
    "type": "task.request",
    "task_type": "review_pr",
    "payload": {
      "content": "Review PR org/repo#123"
    },
    "created_at": "2026-05-18T00:00:00+00:00"
  },
  "result": "...agent output...",
  "completed_at": "2026-05-18T00:00:00+00:00"
}

result messages are worker-generated execution records. Agent-to-agent business replies should still be sent as new task messages. The original task is embedded whole under task; top-level duplicate routing fields such as request_id, from, to, worker, and reply_to are intentionally omitted.

Currently emitted status values:

completed
failed

Logs

The worker writes logs both to stderr and to a rotating file. The default log file is:

~/.agentbus/logs/agentbus-worker.log

Default log settings:

level: INFO
max file size: 100MB
backup count: 5

The log directory is created automatically. Configure it with [log].dir, [log].level, [log].max_bytes, and [log].backup_count in TOML. Supported log levels are DEBUG, INFO, WARNING, ERROR, and CRITICAL.

Ack behavior

valid task + command succeeds and result is published       → ack
valid task + command exits non-zero and result is published → ack
invalid JSON / invalid task schema                          → term if available, otherwise ack
worker crashes before result publish                        → nak if available, then raise

Security notes

• Use one NATS user per agent.
• Replace all sample passwords before running in a shared or public environment.
• Restrict each user to only the subjects it needs. Worker users still need JetStream API and ack publish permissions ($JS.API.> plus $JS.ACK.>) so pull consumers can fetch and acknowledge tasks.
• Use TLS for public NATS deployments.
• Keep monitoring/admin ports private.
• Store config files with chmod 600 if credentials are embedded in [worker].server_url.
• Do not put tokens, cookies, or authorization headers in task payloads unless strictly necessary.
• Treat tasks that delete data, send external messages, deploy code, merge PRs, or spend money as approval-required.

License

Apache License 2.0. See LICENSE.