Agentro — Move your workflow where AI lives

The stone age, and the way out

You're hand-carrying context between humans and AI. Agentro inverts it.

Every team bolts AI onto a workflow built for humans — then spends the rest of their time being the glue. Agentro flips the posture: a cloud workspace designed for humans and AI as peers, from the first commit.

The stone age TODAY

You become the context layer.

Paste tickets, re-explain decisions to a fresh agent. Every session starts from zero — the agent doesn't remember the repo, the decision, the why.

Three windows in your head.

One for the AI. One for your team. One for the code. You're the glue. The glue is exhausted.

Secrets in the blast radius.

Keys live where the agent runs. One leak and you pull them — but then the agent can't do real work.

Agentro's inversion THE WAY OUT

The system orchestrates.

Not you stitching tools. A system composes humans, AI, and tools into one workflow — with humans as the value-judgment layer. Talk to the agent like a teammate.

Context travels with the actor.

When work moves between people and agents, the repo, the decisions, the why — all carried. One continuous thread.

Isolation by default.

Disposable sandbox. Credentials brokered by the platform. Security is structural, not a policy you trust.

See it work

Conversations with an agent in them move at a different speed.

Agentro was born because a non-engineer founder and an engineer founder wanted to ship through an AI agent — together, in the same thread. When AI sits inside the conversation instead of outside it, engineers and non-engineers stop translating and start building. Watch it across teams.

#enterprise-pilot-pricing

Dana Sales lead

We've got a prospect who wants way more credits than our published cap. Can someone model out 5M credits/mo with a blended rate? Need it before the 3pm call.

Ryo Engineer

I can build the gating logic, but I need the tier table first. @agentro, wanna draft both?

Agentro

Plan · volume-tier-blend.md

0–1M · $0.12 / credit

1M–3M · $0.09 / credit

3M–5M · $0.07 / credit (committed)

✓ floor margin holds at 41% (cf. 38% target)

Ryo Engineer

3M+ tier blows our cost ceiling if they go frontier. Can we gate model choice above 3M?

Agentro

PR #211 · feat/volume-tier-gating

+ if (credits > 3_000_000) tier.gateModelSwap = true

✓ 18 tests passed · mentions @ryo for review

Dana Sales lead

Perfect, that's exactly what I need. Taking this to the call. Ryo, let me know when it merges.

#churn-diagnostics-q3

You Founder

Churn is up 3pts this quarter. I think it's the onboarding redesign but I'm not sure. @agentro, pull the cohort data and tell me if I'm wrong.

Noor Data analyst

Hold on — v3 also shipped a pricing change the same week. @agentro, can you segment by acquisition channel too?

Agentro

Cohort retention · v2 vs v3 onboarding

v3 (redesign) W8 retention: 61%

v2 (prev) W8 retention: 69%

Delta concentrated in the "first task" funnel step. Pricing-page effect null (±1pt).

Noor Data analyst

OK so onboarding IS the signal. Let me draft a fix for the first-task step.

You Founder

Alright, not a full rewrite then. Noor, handle the fix. @agentro, open the branch.

Agentro

Branch · fix/onboarding-first-task

+ onboarding_v3: coef −0.081 (p=0.003)

+ pricing_change: coef 0.004 (p=0.71)

#incident-418-upload-timeout

Leo Support

We've got 3 enterprise tickets in 20 min about large uploads timing out on Edge. Logs show 504s at the gateway. I can't reproduce it.

Priya Engineer

Probably edge config. @agentro, pull the gateway config and the upload pipeline, see what's off.

Agentro

Branch · fix/edge-upload-limit

+ client_max_body_size 50m; # was 10m on edge

✓ repro confirmed: 32MB upload now succeeds

Priya Engineer

Nice catch. But 50m everywhere is asking for trouble. Gate it by auth and add streaming for 25MB+.

Agentro

PR #418 · fix/edge-upload-limit

+ if authed: client_max_body_size 50m else 10m

+ streaming path for content_length > 25m

✓ 24 tests passed · mentions @priya for review

Leo Support

I'll let the 3 customers know the timeline. This would've taken us a full day the old way.

#export-csv-feature

Sara Product manager

Users keep asking for CSV export. I wrote a quick spec. @agentro, can you turn this into something shippable without a sprint meeting?

Ken Engineer

One thing — the streaming writer has to respect the row-level ACL from last month. Don't ship something that leaks rows.

Agentro

Plan · csv-export.md

1 · streaming CSV writer (memory-safe, ACL-aware)

2 · export endpoint + queue job

3 · rate limit + download link expiry

Scope: ~1 day of eng. No sprint needed.

Agentro

PR #302 · feat/csv-export

+ for row in acl.filter(queryset, user):

+ writer.write(row)

✓ 31 tests passed · denied-row case green

Sara Product manager

One paragraph to a reviewable PR, all in one thread. This is what I wanted.

#billing-recon-anomaly

Mina Ops

Stripe and our ledger are off by ¥1.2M this month. Doing it by hand is two days. @agentro, take a look?

Jun Data engineer

Join both on charge id. If there's a gap, cluster by time window — usually points to a deploy.

Agentro

Reconciliation · jul-stripe-vs-ledger

Matched: 4,812 charges

Stripe-only: 23 (¥1.18M)

Ledger-only: 1 (¥0.02M)

23 missing charges — all from 14:00–15:00 on Jul 14. Webhook outage window.

Jun Data engineer

Jul 14 2pm — that's the deploy window. Webhook handler wasn't draining on restart. Confirm it and write the backfill plus the fix.

Agentro

PR #509 · fix/webhook-drain-on-restart

+ async onRestart() { await this.queue.drain() }

✓ backfill: 23/23 charges replayed · ledger reconciled

Mina Ops

Two days of manual work → one thread. Closing the books.

What's in the box

A coding agent that ships, and the integrations that let it reach the rest of your stack.

Two layers. The coding agent reads, writes, tests, and merges inside an isolated sandbox. The integration layer connects it to GitHub, your tools, and any MCP-capable surface — so the agent acts where your work already lives.

01 · Coding agentReads code. Writes code. Runs tests. Opens PRs.

Branch, commit, PR

Opens branches, writes commits, opens reviewable PRs — never pushes to main without a human's approve.

Workflow

Repo-aware context

Reads the whole repo, follows conventions, and carries decisions across sessions. No re-explaining the codebase.

Context

Tests & verification

Runs the test suite, interprets failures, and iterates until green — before it ever asks for review.

Quality

Disposable sandbox

Every run lives in an isolated microVM. Throw it away when done. Your host is never touched.

Isolation

Model-agnostic

Swap the brain per task — frontier for hard reasoning, cheap for routine edits. No vendor lock-in.

Models

Inline approval gates

Irreversible actions stop for a human. Approve, reject, or redirect — every gate is logged.

Control

02 · Integrations & MCPReach your whole stack — GitHub, SaaS, and any MCP server.

GitHub native

OAuth in seconds. Branches, PRs, reviews, and merges flow through GitHub's own permissions and branch rules.

Source control

MCP-compatible

Connect any Model Context Protocol server. Bring your own tools, data sources, and capabilities — Agentro speaks the standard.

Protocol

Browser & web

A real browser the agent drives — for sites without APIs, internal apps, and anything behind a login.

Web

SaaS & APIs

Pluggable integrations for the SaaS and APIs your team already uses. Auth handled by the platform, not the agent.

Integrations

Proxy-only credentials

Keys and tokens never reach the agent's sandbox. The platform holds them and brokers each call.

Secrets

Webhooks & events

Trigger agents from issue comments, PR events, or external webhooks. Work starts where the signal is.

Automation

Build log

We run our company on this. Every feature is something we needed.

We're a 2-person startup in Tokyo. We built Agentro because our own team needed it — then cut it loose for everyone else. Every line of this product is dogfooded first.

Shipped practice, not theory

We use Agentro to ship Agentro. This LP, this copy, this pricing model — drafted with the agent you'd be using.

From the frontier

We track the edge of what AI agents can do, because we have to. The insights land in the product, in public.

Open about the tradeoffs

What works, what doesn't, what we're still figuring out. In the build log. In public. No vapor.

Security

We didn't bolt security on. We built it in, from the first commit.

Most agent tools ask you to trust a policy. Agentro's security is structural — the agent physically cannot reach your credentials, because they never sit in the same place. This is the part the founders obsessed over, because they run their own company on it.

Disposable sandbox isolation

Every agent runs inside a disposable microVM. Your code and the agent's execution never share a host. When the task ends, the sandbox is destroyed. There is no persistent surface to compromise.

Credentials never reach the agent

API keys, tokens, and secrets live in a broker the agent cannot read. The platform holds them and proxies each call. A prompt injection, a leaked log, a sandbox escape — none of them surface a credential, because it was never there.

Approval gates on irreversibility

Merge, deploy, send, pay — irreversible actions stop for a human. Every approval is logged with the actor, the action, and the diff. You get the speed of an agent with the control line of a human review.

Auditable by construction

Every step — model call, tool execution, branch, approval — is recorded as a session you can replay. No black boxes. No "trust the agent." You see exactly what happened, in order, after the fact.

Pricing

Start free. Pay when the agent earns its keep.

No seats — invite your whole team for free. Pay per credit, tied to work done. Pick the right model for each task to optimize cost: frontier for hard reasoning, lightweight for routine edits.

Free

Enough monthly agent work to feel the magic.

10 credits / month

1 sandbox at a time

Unlimited human participants

Hobby

$20/ mo

For individuals and small teams getting started with AI agents.

200 credits / month

Shared workspaces

MCP + custom integrations

Model swap per task

Top up credits anytime

Priority support

Pro

$100/ mo

For teams running parallel agents across repos.

1,000 credits / month

Shared workspaces

MCP + custom integrations

Model swap per task

Top up credits anytime

Audit log

Enterprise

Custom

For orgs running fleets of agents with governance needs. Credits based on contract.

Credits set by contract

Shared workspaces

MCP + custom integrations

SSO / SCIM

SLA + dedicated support

* Some Enterprise features are on the roadmap. Contact us for details.

Pricing details finalize before public launch. The shape won't change: pay for the agent's work — never for your people.

Stop bolting AI onto your tools. Move your work where AI already lives.

You're hand-carrying context between humans and AI. Agentro inverts it.

The stone age TODAY

You become the context layer.

Three windows in your head.

Secrets in the blast radius.

Agentro's inversion THE WAY OUT

The system orchestrates.

Context travels with the actor.

Isolation by default.

Conversations with an agent in them move at a different speed.

A coding agent that ships, and the integrations that let it reach the rest of your stack.

Branch, commit, PR

Repo-aware context

Tests & verification

Disposable sandbox

Model-agnostic

Inline approval gates

GitHub native

MCP-compatible

Browser & web

SaaS & APIs

Proxy-only credentials

Webhooks & events

We run our company on this. Every feature is something we needed.

Shipped practice, not theory

From the frontier

Open about the tradeoffs

We didn't bolt security on. We built it in, from the first commit.

Disposable sandbox isolation

Credentials never reach the agent

Approval gates on irreversibility

Auditable by construction

Start free. Pay when the agent earns its keep.

Move your workflow where AI lives.

Stop bolting AI onto your tools. Move your work where AI already lives.

You're hand-carrying context between humans and AI. Agentro inverts it.

The stone age TODAY

You become the context layer.

Three windows in your head.

Secrets in the blast radius.

Agentro's inversion THE WAY OUT

The system orchestrates.

Context travels with the actor.

Isolation by default.

Conversations with an agent in them move at a different speed.

A coding agent that ships, and the integrations that let it reach the rest of your stack.

Branch, commit, PR

Repo-aware context

Tests & verification

Disposable sandbox

Model-agnostic

Inline approval gates

GitHub native

MCP-compatible

Browser & web

SaaS & APIs

Proxy-only credentials

Webhooks & events

We run our company on this. Every feature is something we needed.

Shipped practice, not theory

From the frontier

Open about the tradeoffs

We didn't bolt security on. We built it in, from the first commit.

Disposable sandbox isolation

Credentials never reach the agent

Approval gates on irreversibility

Auditable by construction

Start free. Pay when the agent earns its keep.

Move your workflow where AI lives.

Join the waitlist

You're on the list