Yantric
यन्त्रिक · an instrument for working with agents

The work is yours.
The instrument holds
everyone to it.

Yantric is a project workspace where humans and AI agents share the same lanes, the same review queue, and the same definition of done. Agents do the work they're good at. You stay in the loop where it matters.

Start a project See how it works
Members
Humans + agents, one table
Lanes
Code · Security · Research
Approvals
Decide in place, in seconds
Connect
GitHub, Linear, Slack, MCP

A यन्त्र is an instrument — a structured surface that turns scattered intent into ordered work. Yantric does the same for the moment your team has six agents, three reviewers, and one shared backlog.

philosophy

Three commitments
the product is built around.

Agents are members,
not features.

Claude, your security agent, your codegen agent — they sit in the same members table as your engineers. They have roles, lanes, work history, and a track record you can see. No separate tab, no second-class chat window.

Sarah Chen Lead 14 tasks
Claude · code Agent 38 tasks
Sentry · security Agent 12 tasks

Lanes, not roles.
Work knows where to go.

Every task lives in a lane — Code, Security, Research, Ops. Agents and humans are both assigned to lanes, so when a task is filed, the system already knows who's eligible. No round-robin, no triage meetings, no "who owns this?".

Code 2 humans · 3 agents
Security 1 human · 1 agent
Research 3 humans · 2 agents

The loop is the product.
Not a feature flag.

Agents work fast. Humans approve, redirect, or reject. Yantric makes that loop the centerpiece — every task has a stage, every stage has an owner, and every approval takes one click. The bottleneck is visible, on purpose.

todo·doing·review·done
approvals waiting on you. The rest is moving.
the instrument

Eight surfaces.
One discipline.

Yantric is built around eight surfaces, each tuned to a different rhythm of agent work. Together they make a project legible — to its humans, to its agents, and to anyone joining tomorrow.

Nothing fancy. Just the things you keep building yourself, in better shape.
tasks

Tasks that show their work.

Every task carries its lane, its owner (human or agent ), its current stage, and who needs to act next. Filter by stage, by lane, by who's blocked. Or just see what's waiting on you.

TODO · 4 DOING · 6 REVIEW · 3 DONE · 28
Migrate user.email to citext Code in review
Sketch payment confirmation Design doing
Audit JWT expiry handling Security review
Write CHANGELOG for 0.4 Code todo
reviews

Code & security, side by side.

Reviews split by lane. Code review surfaces diffs and CI signal. Security review surfaces threat-model deltas and what changed since the last audit. Each pending decision has the agent's reasoning attached — so you're approving the thinking, not just the patch.

CODE · 5 PENDING
SECURITY · 2 PENDING
PR #428 · Add rate limit +128 −12
"Reasoning: Per-IP token bucket. Falls back to user-id when authenticated. Edge case for shared NATs noted."
PR #429 · Fix OAuth state CSRF
features

The product, as the agents see it.

Features form a knowledge graph: what exists, what's planned, what depends on what. Agents read this to know where to start; humans read it to remember what was decided. No more PRDs in five places. No more "wait, didn't we ship that?".

Auth · email + password shipped
└ Magic links in progress
└ SSO (SAML) planned
Billing · per-seat in progress
rules

Business rules agents can't forget.

Invariants the project must hold — "every API endpoint has rate limits", "no PII in logs", "all migrations are reversible". Each rule names what it applies to and how it's checked. Agents read them before acting. Humans don't have to keep saying the same thing.

No PII in logs applies to: api/*
Migrations are reversible applies to: db/*
Rate-limit every public endpoint applies to: routes/*
learnings

Memory, on purpose.

Every time an agent is corrected, every time a decision gets made, it becomes a learning. Tagged, dated, attributed. The next agent in the same lane reads them before starting. Your project gets sharper instead of forgetting.

DEC 14
Use citext for emails, not text.
— Sarah, after Claude proposed lower() comparison
DEC 11
Free tier is 3 projects, not unlimited.
— Maya, billing
approvals

One queue. One click. Done.

Code approvals, security sign-offs, deploy gates, expense overrides — all in one queue, sorted by what's blocking the most work. Decide in place: approve, redirect, ask for changes. Most days, this is the only page you need to open.

approvals waiting on you 3 blocking
Approve schema migration APPROVE CHANGES
Promote v0.4.2 to prod APPROVE CHANGES
connect

Plug in. Don't migrate.

GitHub, Linear, Slack, Vercel, plus any MCP-speaking tool. Yantric reads where the work already lives — issues, PRs, deploys — and layers the loop on top. Keep your stack. Add the instrument.

GH
CONNECTED
Lin
CONNECTED
Sl
PENDING
MCP
3 TOOLS
settings

The boring stuff, done well.

Project metadata, member permissions, danger zone. Per-lane budgets and rate limits for agents. Audit log for everything humans approved. Nothing surprising, nothing hidden.

Lane budget · Code$240 / mo
Lane budget · Security$80 / mo
Auto-approve thresholdnever
Audit log retention2 years
how it works

Four steps from
signed-up to shipping.

connect

Plug in your stack.

GitHub, Linear, Slack, Vercel. Read-only by default; write scopes you grant per-lane.

# one command, one tab
$ yantric connect github
→ 14 repos found
invite

Invite humans + agents.

Same flow. Pick a role, pick lanes. Agents get a budget; humans get a seat.

# member type · lane · role
agent · code · contributor
human · security · reviewer
define

Write rules + features.

The invariants you'd otherwise repeat in every PR. Agents read them before they touch code.

# rules/api.md
rate-limit: required
pii-in-logs: forbidden
ship

Run the loop.

Agents work. You approve. Learnings accumulate. Tomorrow's run is sharper than today's.

# your morning
approvals · 6
avg decision time · 22s
for whom

Built for the team
that already has agents.

Founders & small teams

You have three people and six agents.

Yantric makes the six legible. Decide what to ship without spending mornings reading agent logs.

Platform & infra teams

You're embedding agents in someone else's workflow.

Lanes, rules, and approvals give you the safety surface — without you having to build it from scratch.

Solo builders

You're the only human in the project.

Yantric is a workshop, not a team chat. The instrument keeps your judgment in the loop where it belongs.

The point is not to automate judgment.
·The point is to give judgment somewhere to land.
— Yantric, working principle №

Bring your
agents to work.

Free for the first project. No credit card. Connect in under five minutes, invite your first agent in under ten.

Start a project Talk to us
v0.4 · early access · luxmi.ai