Example query: SELECT * FROM "orders.v2" WHERE value.status = 'FAILED', which returns 1,284 rows in 4.2 seconds.

Kafka, queried like a database. Read-only until you say otherwise.

For engineers who debug Kafka in production. So, all of you.

Download for macOS Windows · LinuxPUBLIC BETA

NegeseuonWRITES LOCKED

Query · failed ordersorders.v2Consumer groups

SELECT partition, offset, key, value.status, value.reason, timestampFROM "orders.v2"WHERE value.status = 'FAILED' AND timestamp > now() - INTERVAL '1 hour'ORDER BY timestamp DESC;

scanning 12 partitions312 rows · 61 MB · 1.3s

partition	offset	key	value.status	value.reason	timestamp
7	8412311	ord-77ab10	FAILED	card_expired	14:02:11.418
3	8412034	ord-8c92f4	FAILED	payment_declined	14:02:09.903
1	8411950	ord-1f03cc	FAILED	insufficient_funds	14:01:58.240
10	8411876	ord-a20d7e	FAILED	3ds_timeout	14:01:51.007
3	8411744	ord-c4e981	FAILED	payment_declined	14:01:40.662
5	8411702	ord-5b6e42	FAILED	payment_declined	14:01:33.115
9	8411688	ord-e1c3b9	FAILED	address_invalid	14:01:29.874
0	8411521	ord-90fd21	FAILED	card_expired	14:01:12.330
6	8411490	ord-3a7c55	FAILED	payment_declined	14:01:08.951
11	8411406	ord-d84b02	FAILED	3ds_timeout	14:00:57.204

partition

offset

key

value.status

value.reason

timestamp

8412311

ord-77ab10

FAILED

card_expired

14:02:11.418

8412034

ord-8c92f4

FAILED

payment_declined

14:02:09.903

8411950

ord-1f03cc

FAILED

insufficient_funds

14:01:58.240

8411876

ord-a20d7e

FAILED

3ds_timeout

14:01:51.007

8411744

ord-c4e981

FAILED

payment_declined

14:01:40.662

8411702

ord-5b6e42

FAILED

payment_declined

14:01:33.115

8411688

ord-e1c3b9

FAILED

address_invalid

14:01:29.874

8411521

ord-90fd21

FAILED

card_expired

14:01:12.330

8411490

ord-3a7c55

FAILED

payment_declined

14:01:08.951

8411406

ord-d84b02

FAILED

3ds_timeout

14:00:57.204

orders-value · AVRO · REGISTRYWrites lockedp3 · 8412034

ord-8c92f4p3 · offset 8412034 · 412 B · Avro

{
  "order_id": "ord-8c92f4",
  "status": "FAILED",
  "reason": "payment_declined",
  "amount": 129.40,
  "currency": "GBP",
  "attempt": 3,
  "customer": {
    "id": "cus-41d2",
    "tier": "gold"
  },
  "items": [
    { "sku": "SKU-1180", "qty": 2 },
    { "sku": "SKU-0042", "qty": 1 }
  ],
  "psp": "adyen"
}

Your Kafka UI vs. Negeseuon

Loads 100 messages a page

Scrolls the whole topic, both ways, in flat memory

Can't search inside payloads

WHERE value.anything = 'you want'

Creates consumer groups, commits offsets

Joins no group. Commits nothing. Leaves no trace.

One misclick from deleting prod

Writes locked until you unlock them, enforced in the Rust core

Schema? Good luck.

Avro, Protobuf, JSON Schema, decoded on sight

SQL over topics

Ask the topic a question. Get rows back.

Reach into decoded payloads with value.customer.id, headers with headers['trace-id']. Filters stream as you scroll; aggregates, sorts and window functions run in DuckDB. Explain shows exactly what gets read, and a key lookup only reads the partition that key lives in.

QueryPRODprod-eu-westPRO · JOIN

SELECT o.key, o.value.customer_id, o.value.total, p.value.reasonFROM "orders.v2" oJOIN "payments.events" p ON p.value.order_id = o.keyWHERE o.value.status = 'FAILED' AND o.value.total > 50 AND o.timestamp > now() - INTERVAL '6 hours'ORDER BY o.value.total DESC;

Scanning orders.v2 · 12 partitions10 of 12 done

p10

p11

1,284 rows · 258 MB · 4.2s

o.key	customer_id	total	p.reason	partition	offset
ord-1f03cc	cus-93aa	312.99	insufficient_funds	1	8409552
ord-5b6e42	cus-2ad8	149.00	payment_declined	5	8408440
ord-8c92f4	cus-41d2	129.40	payment_declined	3	8412034
ord-a20d7e	cus-7f63	76.25	3ds_timeout	10	8408907
ord-77ab10	cus-0b9e	54.00	card_expired	7	8411876

PROJoins and queries across topics, plus export to CSV, JSON Lines or Parquet, saved queries and history. SQL over a single topic is free.

Keyboard-first

⌘P. That's the interface.

failTopics · Tabs · Queries · History

Saved queries

Failed orders, last hourprod-eu-west

Failed payments by reasonprod-eu-west

Topics

payments.failures.dlqstaging

Commands

Unlock Writes…⇧⌘U

Run Without Cache⇧⌘↵

⌘P

Find anything. Topics on every connection, open tabs, saved queries, history. Type select and it becomes a SQL editor.

⇧⌘P

Every command, by name. If it doesn't apply here, it says why.

⌘↵

Run the query. Esc cancels it, mid-scan.

⇧⌘U

Unlock writes. On purpose, with a timer.

Speed, without side effects

Zero consumer groups.Zero offsets.Zero apologies.

0consumer groups created

0offsets committed

Flatmemory, at any topic size

Δre-runs read only what's new

Scan Inspectororders.v212 partitions48,213,907 recordsMB/s per partition

Click any progress bar to open it. Per-partition rate and ETA, query memory against its limit, and every record that failed to decode, with the reason.

Schema-aware

Whatever's on the wire, decoded.

Every message goes down the chain until something can read it. You never pick a deserializer from a dropdown, though you can pin one per topic when you want to.

Registrymatched

magic 0x00 · schema id 14 · Avro

{ "order_id": "ord-8c92f4",
  "status": "FAILED",
  "amount": 129.40,
  "currency": "GBP" }

JSONfallback

parses as JSON

{"order_id":"ord-8c92f4","status":"FAILED","amount":129.4}

UTF-8fallback

valid text

ord-8c92f4|FAILED|129.40|GBP

Hexlast resort

always works

00 00 00 00 0e 14 6f 72
64 2d 38 63 39 32 66 34
0c 46 41 49 4c 45 44 …

Schemas from

Your registry · a watched folder · a descriptor set

Formats

Avro · Protobuf · JSON Schema · MessagePack · UUID ·ints and floats

Fetched schemas are cached per connection, so framed records still decode when the registry is down. Edit a .proto in a linked folder and it's picked up within a second.

Production safety

Read-only. Until you mean it.

01Mark a workspace Production. It gets a red badge and refuses every write, on every launch.
02Writes need a time-limited unlock: 15 minutes or an hour, never longer in production. The unlock lives in memory, so a restart starts locked.
03Every send to production is confirmed, naming the workspace, connection, topic and count. Enforced in the Rust core, not by greyed-out buttons.
04Deleting a topic means typing its name. All of it. orders.v2, not "yes". Internal topics can't be touched at all.

Unlock writesPROD

prod-eu-west refuses writes. Unlocking lets you produce, create and delete topics until the time is up. Production workspaces unlock for at most 1 hour.

15 minutesuntil 14:17

1 houruntil 15:02

CancelUnlock for 15 minutes

Type orders.v2 to delete it

orders.v

AI agents · MCP

Your agent is welcome. Your customers' emails aren't.

Connect your agent with one click. Negeseuon finds the agents on your machine and writes itself into their settings. Twelve tools, all read-only, run through the app's own code over a socket only you can open. No port, no cloud relay.

Claude CodeClaude DesktopCursorVS CodeCodexany MCP client

order_idcustomer_emailcard_numberstatustotal

ord-90fd21nx_3b71c0e9a4…withheldPAID42.10

Claude Code is reading

ord-8c92f4nx_8f2c51d07b…withheldFAILED129.40

ord-77ab10nx_19ae6f2c38…withheldFAILED54.00

ord-1f03ccnx_c07d93ab12…withheldFAILED312.99

ord-e1c3b9nx_8f2c51d07b…withheldSHIPPED18.50

Tokens are the same for the same value within a session, so GROUP BY and joins still work. They change every session, and nothing maps them back.

Agent access · prod-eu-west

Off

Structure onlynames, counts, schemas; no values

Chosen fieldsshow, token or withhold

All contents

What agents read

14:02:11  Claude Code  query_run   orders.v2
          40 rows · 80 tokens · 40 withheld
14:02:14  Claude Code  schema_get  orders-value
          Names and counts

Every call that sent an agent anything, by session. Names and counts only, never a value. Production asks you first, every session.

And you watch it work. The window glows violet while an agent is busy, a pointer with its name moves to whatever it touches, and its SQL types itself out in a tab of its own. Stop ends the session.

And the everyday stuff.

Live tail: Follow a topic as it's written. Filter it, pause it, see how many arrived while you looked away. Memory stays capped however busy the topic is.
Consumer groups and lag: Every group, member and partition, and how far behind. Matches rpk group describe. Read from the outside, never joined.
Produce PRO: One message, a JSON Lines file of up to 100,000, a saved template, or an exact copy of a record you already have. Schema-encoded, with a generated example to start from.
Share links: negeseuon:// links open the same topic, record or query in a teammate's app, on their connection. No hosts, secrets or payloads in the link.
Any sign-in: SASL PLAIN, SCRAM, OAUTHBEARER, AWS MSK IAM and mTLS. Or import a client.properties and the password goes straight to the keychain.
Signed auto-updates: Stable and Beta channels. Every update is signature-checked before it installs, and switching channel never downgrades you.

Kafka · Redpanda today.MQTT · RabbitMQ · BullMQ next.

We never see your messages. Not even the boring ones.

Secrets stay in the OS keychain.

Passwords, keys and tokens live in macOS Keychain, Windows Credential Manager or the Secret Service on Linux. Never in the database, logs or error messages.

Caches are encrypted.

Query sessions are encrypted at rest and removed on close. Nothing from a message ever reaches a log file.

No telemetry. No analytics.

The app talks to your brokers, your registries and our update server. That's the list. Read it in full.

Free to look. Pro to touch.

Free

£0forever

Look at anything. Touch nothing.

Browse, decode and SQL over one topic at a time
Live tail, consumer groups and lag
Every sign-in: SASL, SCRAM, OAuth, AWS MSK IAM, mTLS
Command palette, share links, signed updates
Workspaces and connections: limit set at launch

Pro

Monthlyprice set at launch

Everything in Free, plus the parts that change things.

Queries across topics: joins, the Query tab, export to CSV, JSON Lines, Parquet
Produce, re-produce, bulk send, templates; create and delete topics
AI agents over MCP, with per-field access rules
Cached re-runs, saved queries and history
Unlimited workspaces and connections

Enterprise

Talkto us

Pro for every seat, plus control over all of them.

Offline licence that never phones home
Managed policy and version pinning
Telemetry off, or sent to your own collector
Credentials from your secret store

During the beta, every feature is unlocked.Compare every feature →