Read nothing.

Observability for the new way of working.

Elk reads how your team, your systems, and your product are doing, then hands the next move to your team and agents. Taste and judgment stay yours.

Agents doubled the output. Nobody doubled the reading.

Every coding session, pull request, call, deploy, and error is exhaust from work that already happened. It is the richest signal a team has and the least watched. Reading it stopped scaling. Watching it never started.

Activity everywhere. Observation nowhere.

Three signals.
One reader.

Elk watches the three things a company gives off while it works, and reads them together, because a slow checkout, a hot pull request, and a quiet customer are usually the same story.

Collaboration signal
agent sessions, pull requests, calls, chat
Systems signal
errors, traces, deploys, logs, uptime
Customer signal
usage, retention, what people hit
▸▸▸
Elk Scout
▸▸▸
Lookouts
watches that fire, with the evidence
Elk Tasks
contextual and skilled, optimized for agents
Camps
the work on one thing, in one place
Tracks
the day's read, per person
COLLABORATION SIGNAL

How the building is going

Agent sessions, pull requests, calls, and chat. Elk reads what the team and its agents did, not what they said they would do. This is how we run Elk itself, every day.
Pull requests, by authorbreakdown
36merged this week · +40% on the week before
Claude14
Maya9
Priya6
Cursor3
4 others4
GitHub · last 7 days · 36 rows · read 4m ago
Agents opened four in ten pull requests this week. The review queue, not the writing, is the slow step.
Wired today
MCP (Claude, ChatGPT, Gemini), GitHub, Google Meet, Bugle (Elk Meet), Streams (Elk channels).
SYSTEMS SIGNAL

How it is running

Errors, traces, deploys, logs, and uptime, tied back to the change that caused them. When something needs a call you get told, with the picture attached. There is no wall of graphs to check on a good day.
Checkout errorstrend
4.1%×3 since the 14:20 deploy
Sentry · last 24 hours · 41 events · the dashed rule is the trailing week, which is what normal means
Error rate climbed after the 14:20 deploy and has not come back. Roll back, or hold?
Wired today
Sentry, AWS CloudWatch, Netlify, Cloudflare, Neon, Stride (Elk traces), Catch (Elk errors).
CUSTOMER SIGNAL

How it is landing

Who is using what, what they hit, and whether they come back. Elk brings you the pattern and who it is happening to, not a funnel to interpret.
Weekly retentioncohort
38%still here in week 4, August 25 cohort
1234 8/1151%31%24%19% 8/1854%33%26%21% 8/2571%55%44%38% 9/168%51%
PostHog · 4 cohorts · a dash is under 25 people or not yet elapsed
The August 25 cohort holds better at every week. That is the onboarding change landing.
Wired today
PostHog, Trail (Elk analytics).

Every connector is early and growing. Don't see yours? Name it in the app and it goes on the list.

Every signal ends in a next move.

A person, if it needs taste. An agent, if it needs hands.

Elk turns what it saw into one concrete task, with the evidence attached, and hands it to whoever should do it. The loop closes when the pull request merges, and Elk sees that too.

Evidence attached
The commits, the alert, the call, and the session that led here, on the task itself.
Handed to a person or an agent
Taste goes to a colleague. Hands go to an agent. Either way, Elk keeps watching.
Closed by the work itself
A merged pull request closes the task. Nobody updates a board on Friday.
Elk task · from Claude's finding
Add a circuit breaker to the payment webhook path
Webhook timeouts traced to a provider retry storm. Two pull requests, one Sentry issue, and one call attached.
Evidence
#1284 retry with backoff merged into mainGitHub · 38m
PaymentWebhookTimeout 41 events since the 14:20 deploySentry · 4m
Payments sync “add that breaker Claude mentioned”Bugle · 12 min
Demo data

Camps form on their own.

One thing. One place.

When conversations, commits, alerts, and usage start pointing at the same thing, Elk names it and gathers the evidence in one place. A camp is where the work on one thing lives: what changed, who is on it, what is watched, what is next.

Forming · last 36 hours
Payments reliability
Gathered from conversations, code, monitoring, and usage. One incident, now resolved.
4
conversations
3
agents involved
4·2
PRs · commits
2
Lookouts
Tasks
Add a circuit breaker to the payment webhook pathfrom Claude's finding
Add retry + backoff on transient provider errors2 PRs reference this
Update the runbook with the new failure modeflagged in the incident
Demo data

Built for the person who cannot read it all.

Skim less. Know more.

Six questions every leader asks. Elk answers them from what is actually happening, and tells you when the answer changes.

01What are we building?
02Who and what is producing it?
03Are we on track?
04Is anyone using it?
05What is it costing?
06What is about to go wrong?
Usage & insights
Real behavior
12,543+18%
active users this month
Camps92%
Voice64%
Ask Elk51%
4.6/ 5 experience score
Cost visibility
Agents + infra
$34,870this month
Claude52%
GPT-4o26%
Other22%
AWS infra$12.1k
GCP infra$6.4k
Lookouts
3 watching · 1 fired this week
Checkout error rate climbed after the 14:20 deploy
Fired · 4m ago · roll back, or hold?
Runbook update untouched for 7 days
Fired · Tuesday · owner confirmed
Sessions that end without a pull request
Watching · quiet
Weekly active users down more than 5%
Watching · quiet
You do not watch the graph. The graph watches for you.

Other systems make the data. Other systems do the work.
Elk is the middle.

Five verbs, one loop, and it runs continuously.

01
Observe
What the team, the systems, and the customers give off.
02
Distill
What changed, and what it means.
03
Judge
A person decides what matters. Elk shows the evidence.
04
Guide
The next move, with the context attached.
05
Do
A person or an agent, and Elk sees it land.

Pilot small.
Expand as it earns it.

Start with the team already running agents.

Elk lands with the team already running agents, already on call, already shipping faster than it can read. Connect one signal from each of the three and let it watch for a week. No big-bang rollout.

Where permission exists, change begins.

Companies don't need more dashboards. They need to see themselves work.

Read nothing. See everything that matters.

Invite-only, opening in waves. No spam. Just your invite when it's ready.