Use our RESTful API in your own stack, the CLI to set up locally, or the MerkleBot Agent on your robots. Here's how the pieces fit.
Install the CLI, register a device and create a pipeline. The agent handles recording, compression, encryption, hashing and resumable uploads.
Examples on this page are illustrative. Full API reference and SDK credentials are available in your workspace after sign-up.
# 1. Install the CLI and log in
pip install merklebot-cli
mb login
# 2. Register a robot and install the agent
mb devices add --name ur5e-cell-03 --site warehouse-b
mb agent install --device ur5e-cell-03
# 3. Record ROS 2 topics to MCAP and sync to hybrid storage
mb pipelines create --device ur5e-cell-03 \
--source ros2 --topics "/camera/.*,/joint_states,/tf" \
--format mcap --compression zstd --storage hybrid# Query recordings with the REST API
curl -s https://api.merklebot.com/v1/recordings \
-H "Authorization: Bearer $MB_TOKEN" \
-G --data-urlencode "device=ur5e-cell-03" \
--data-urlencode "from=2026-10-01T00:00:00Z" \
--data-urlencode "event=gripper_stall"from merklebot import Client
mb = Client(token="MB_TOKEN")
# Run a pre-built extractor on every new bag from a device
mb.extractors.attach(
device="ur5e-cell-03",
extractor="rosbag-video",
params={"topic": "/camera/color/image_raw", "fps": 15},
)
for rec in mb.recordings.search(event="gripper_stall", last="7d"):
print(rec.id, rec.cid, rec.video_url)One YAML file describes sources, triggers and upload rules. Record continuously at low rate, and capture full-rate snapshots around events — before and after the error.
# /etc/merklebot/agent.yaml
device: ur5e-cell-03
site: warehouse-b
sources:
- type: ros2
topics: ["/camera/color/image_raw", "/joint_states", "/tf"]
storage: mcap
compression: zstd
split: { max_size_mb: 1024, max_duration_s: 300 }
- type: mqtt
broker: tcp://gateway.local:1883
topics: ["plant/line2/+/telemetry"]
triggers:
- on: "/diagnostics level>=ERROR"
snapshot: { before_s: 60, after_s: 30 }
upload:
schedule: "idle_or_charging"
max_bandwidth_mbps: 20
encryption: client-side
seal: merkle-root # sign each batch with the device keyEvery batch is sealed with a Merkle root and signed by the device key. Subscribe to recording.sealed, extractor.completed or device.offline events and trigger your own workflows.
{
"event": "recording.sealed",
"device": "ur5e-cell-03",
"recording_id": "rec_01J9X4N8T2",
"cid": "bafybeigdyrzt5sfp7udm7hu76uh7y26nf3efuylqabf3oclgtqy55fbzdi",
"merkle_root": "9f3ac1e04b7d2a58c6...",
"signature": "ed25519:4k1Q...",
"size_bytes": 734003200,
"tags": ["gripper_stall", "shift-2"]
}Package your processing code as a container and attach it to devices or events. Outputs are versioned and written back to the index.
# Dockerfile for a custom extractor
FROM python:3.12-slim
RUN pip install --no-cache-dir mcap mcap-ros2-support pyarrow
COPY extract.py /app/extract.py
ENTRYPOINT ["python", "/app/extract.py"]
# register it
mb extractors register --name joints-to-parquet --image registry.example.com/joints:1.2
mb extractors attach --device ur5e-cell-03 --extractor joints-to-parquet --on recording.sealed
| Tool | Use it for |
|---|---|
| REST API | Devices, recordings, search, extractors, webhooks |
| CLI | Local setup, scripting, CI pipelines |
| Python SDK | Notebooks, data science, training pipelines |
| MerkleBot Agent | On-robot recording, buffering, encryption, upload |
| Extractors | ROSbag/MCAP → video, images, Parquet, enriched logs |
Why logging breaks at fleet scale, how to configure rosbag2 with MCAP, compression and snapshot mode, and a reference pipeline from robot to searchable archive.
Read articleHashing, inclusion and consistency proofs, signing roots with device keys and anchoring them in time — the building blocks of verifiable machine data.
Read articleULog flight logs, MAVLink streams and the PX4–ROS 2 bridge: building a pipeline that turns every landing into searchable, analyzable flight data.
Read articleStart on the free Starter plan or talk to an engineer about your stack.
