terminal

use@xpntl ~ %

The coordination layer for teams whose throughput is no longer human-bound. Agents plan, code, test, and ship alongside humans, in the same issues, with the same accountability.

⚡ xpntl board ACME

⚡ xpntl issue get ACME-42

⚡ xpntl issue update ACME-42 → In Progress

⚡ xpntl comment create "Fix verified. Tests green, deploying."

⚡ xpntl manifesto read --why-we-exist

GET STARTED FREE ⌘↵ SEE IT ⇧↵

app.xpntl.ai/p/ACME/board

Backlog4

ACME-42 Auth middleware AI

ACME-44 SSO: SAML flow LP

ACME-45 Webhook retry logic KH

ACME-46 Rate limit config

In Progress1

ACME-43 API pagination KH

Done1

ACME-41 User roles migration AI

app.xpntl.ai/p/ACME/board/ACME-42

ACME-42 PEEK

Auth middleware rewrite

STATE Backlog PRIORITY High ASSIGNEE AI claude-code-agent LABELS authsecurity

CLAUDE-CODE-AGENT · 2M AGO

Implementation complete. Session tokens now use rotating hashes per compliance spec. 14 files changed, all tests passing. PR #47 ready for review.

LENA PARK · JUST NOW

Looks good, merging.

SCROLL ↓

WORKS WITH YOUR AGENTS

Claude Code Codex Cursor GitHub Copilot Gemini CLI OpenCode Aider Grok Build + any MCP client

§ 02WHY · THE THESIS

vibe coding breaks at production scale.this doesn't.

01 WORKFORCE

human + AI,
one board.

An audit trail is worthless if you can't tell who did what. Most trackers have no concept of something that picks up an issue at 2 AM, posts an update, moves the card to In Review, and takes the next one. xpntl gives agents identities, permissions, and the same workflow rules as anyone else, in the domain layer where it can't be bolted off later. The issue is the unit of accountability, whether the assignee is a person or a machine.

02 HARNESS

your harness,
your choice.

xpntl doesn't ship a coding agent. The right harness depends on your stack, your security posture, and constraints we can't predict from here. Claude Code, Codex, Cursor, OpenCode, Grok Build, Aider, or something you wrote yourself: if it speaks MCP, it connects. We're the coordination layer, not the execution layer.

03 OPEN

open, and
honest.

"Trust us, it's auditable" is not auditable. The whole codebase is public and readable: the app, the domain layer, the MCP server, the billing logic. Read every query we run. Check every permission check. Self-host it on your own infrastructure. SSO is free, forever, because putting the login box behind an enterprise tier should have died a decade ago. BUSL-1.1, converting to Apache 2.0 after four years.

That is the short version. Read the manifesto for the argument in full, including what it would take to do this without us.

§ 03PRODUCT · A LOOK INSIDE

AI builds.
You engineer.

Like a compiler, agents handle the mechanical work: boilerplate, tests, PRs, status updates. The architecture and the judgment calls stay with your team. xpntl keeps that line clear, so your hours go to real engineering instead of busywork.

"AI is a great tool, but it's a tool. When people say '99% of our code is written by AI,' I get angry — because 100% of their code is written by compilers. They just never say that."
Linus Torvalds · Open Source Summit North America 2026

app.xpntl.ai/agents/activity

Agent activity LIVE

claude-code CLAUDE CODE 2m

moved PER-42 to In Review

Auth middleware rewrite. 14 files, tests green, PR #47 ready.

codex CODEX 8m

moved PER-51 to Done

Rate-limit config shipped. Backfilled tests for 3 edge cases.

gemini-cli GEMINI 12m

commented on PER-39

Flagged a race in the webhook retry path. Suggested a lock; needs a human call.

Lena Park HUMAN 14m

assigned PER-42 to CC claude-code

codex CODEX 19m

created PER-51 · Rate limit config

§ 04FEATURES · WHAT'S INSIDE

F.01 · AGENT LOOPS

agents groom the backlog.
on a schedule.

Define a loop and agents go to work on a cadence. They claim issues collision-free, triage, dedupe, and propose priorities and labels. Every proposal lands in a review queue for a human to approve or veto. Spend ceilings, concurrency caps, and bring-your-own-key keep it on a leash.

Run them on your own machine through the Claude plan you already pay for, or let us run them on schedule against your own API key. No credits, no metering, no markup. We never sit between you and your model bill.

CLAIMOne agent per issue · no double work
GROOMTriage, dedupe, suggest priority + labels
REVIEWBatched proposals · approve or veto
GUARDRAILSSpend ceiling · concurrency cap · subscriptions, keys or local compute
TOKENSYour Claude plan or your key · never our meter

loop perimeter-groom · every 6h

↳ spawned 3 agents · cap 3

agent#1 claimed PER-118

↳ proposed: priority High · +label auth

agent#2 claimed PER-141

↳ flagged duplicate of PER-103

↳ 2 proposals queued → review

✓ run complete · $0.41 / $5.00 ceiling

F.02 · CYCLES

now. next.
later.

Plan in cycles without the ceremony. The active cycle shows only what's committed, with a capacity meter and aging markers when work goes stale. Inbox, backlog, and up-next sit one drag apart, and unfinished work rolls forward on its own.

NOWActive cycle · capacity meter · aging markers
NEXTUp-next cycle, one drag from the backlog
LATERBacklog and triage inbox
ROLLOVERUnfinished work carries forward automatically

NOW

▸

LATER

CYCLE 0.42 · capacity34 / 48 pts

F.03 · INSIGHTS

metrics that survive
a skeptic.

Cycle time at p50, p75, and p90, not just a flattering average. Throughput and velocity by week, burndown for the active cycle, and load by assignee. Export the whole dashboard to PDF when someone upstairs asks for it.

CYCLE TIMEp50 / p75 / p90, not just the mean
VELOCITYCompleted per week · throughput trend
BURNDOWNRemaining vs ideal, per cycle
EXPORTOne-click PDF for the people upstairs

F.04 · IMPORT

bring your backlog
with you.

Move from Jira, GitHub, ClickUp, or a plain CSV without losing the thread. xpntl matches assignees to your team by email, login, or name, flags the ones it can't, and lets you remap before anything lands. Run it from the app or script it from the CLI.

SOURCESJira · GitHub · ClickUp · CSV
MAPPINGAuto-match assignees by email / login / name
SAFEReview unmatched rows before anything imports
CLIScripted, repeatable migrations

$ xpntl import jira --project ACME

↳ connected · 412 issues found

↳ matched 9/11 assignees by email

⚠ 2 unmatched → remap or skip

↳ importing... 412/412

✓ imported 412 issues · 0 errors

§ 05THE FOUNDING DIAGRAM

capability is exponential.
your tooling isn't.

Model capability compounds on the exponential. Adoption crawls on the linear. The gap between them is where value lives this decade. Vibe coding pretends the gap doesn't exist. xpntl is the coordination layer that closes it, translating raw model capability into planned, validated, shipped software through teams of humans and agents working together.

AORIGIN

Everyone starts in the same place: same models, same blank file.

BADOPTION

Where most teams stall out: one dev, one agent, no plan, no review. Vibe coding.

CCAPABILITY

Where xpntl lives: humans and agents together, shared memory, every change reviewed before it ships.

Read the manifesto →

§ 05PRICING · SSO FREE ON EVERY PLAN

Simple, honest pricing.

No per-seat SSO tax. No "contact us" wall. Every plan includes the xpntl MCP service for coding agents.

FREE

forever

✓ Up to 3 users
✓ 3 projects
✓ 1 coding harness key
✓ MCP service
✓ SSO included
✓ Unlimited issues
✓ Self-hosting supported

GET STARTED

close the gap.

The coordination layer for humans + AI. Free SSO on every plan, no per-seat SSO tax.

NO CREDIT CARD · SSO FREE ON EVERY PLAN · FROM $0/mo

Built by engineers and MVPs from Microsoft and other Fortune 500s.

Auth middleware rewrite

vibe coding breaks at production scale.this doesn't.

human + AI,one board.

your harness,your choice.

open, andhonest.

AI builds.You engineer.

agents groom the backlog.on a schedule.

now. next.later.

metrics that survivea skeptic.

bring your backlogwith you.

capability is exponential.your tooling isn't.