Gates that don’t drift
TDD, acceptance criteria, and review threads are enforced as CI checks and agent FAIL conditions — not polite suggestions in a prompt. Prompts drift. Gates don’t.
AI agents that ship production code — and prove it works.
Otta is an opinionated, self-hostable AI dev pipeline. Every change is test-driven, visually verified, and gated on acceptance criteria before it merges — with a memory layer your agents actually keep.
Proven shipping a real production SaaS — last 90 days
457
merged PRs
last 90 days
440
production deploys
GitHub deployments
7 min
median PR cycle
open → merged
2.8 min
median CI loop
feedback speed
210
semver releases
auto-tagged
6 / 3
blockers caught
on 3 CI-green PRs
Prompts drift. Gates don’t.
Otta doesn’t bet quality on a clever prompt or a bigger model. The rules that keep code correct — a failing test first, a screenshot of every state, acceptance criteria that map to a real check — are wired into the pipeline as gates that can’t be skipped, no matter which agent or model does the work.
The pipeline
Intent in. Verified release out.
Six stages, each with a gate that can’t be skipped. Most tools show you an agent that writes code — this shows you what fires at every step to make sure that code is correct.
- 01
Describe
Plain-language intent becomes a fully-specified issue.
Say what you want. The system writes a Linear issue with a GIVEN/WHEN/THEN acceptance block, estimate, priority, and milestone — no ticket-grooming ceremony.
Acceptance criteria exist before any code does. - 02
Isolate
One agent, one git worktree, zero collisions.
Every issue builds in its own worktree on its own dev slot. Many issues run in parallel without stepping on each other.
Parallel work that never corrupts a shared branch. - 03
Build · test-first
The failing test is written before the code.
The builder writes the smallest red test, makes it green, then typechecks and opens a PR. Typecheck is not test coverage — and the pipeline knows the difference.
No production change ships without a test that proves it. - 04
Review · prove it
Runs the tests, opens the browser, checks the criteria.
The reviewer re-runs the focused tests, browser-verifies the happy path plus loading / empty / error states, and rejects any PR whose acceptance boxes lack real evidence.
Visual + behavioral proof, not “it compiles.” - 05
Ship · health-gated
Serialized merge, verified live before the next one.
Merges hit main one at a time. The deploy waits until the live health endpoint reports the exact merged commit before anything else moves.
The thing that deployed is the thing you reviewed. - 06
Release · zero ritual
Merge → semver tag → GitHub Release → release ledger.
The next version is computed from commit prefixes, tagged, and turned into a GitHub Release and a Linear release entry automatically.
A complete, named release history with no manual steps.
Why it’s different
Autonomy you can actually trust.
Visual verification, built in
User-facing work isn’t done until a real screenshot of the happy, loading, empty, and error states has been checked. A DOM measurement is not a picture.
Memory that compounds
A git-canonical brain plus a cross-agent semantic recall bus means your agents remember decisions, gotchas, and dead-ends — across sessions and across tools.
Self-host the whole thing
Built on Jean (Apache-2.0). Your repos, your infra, your data — by default, on every tier. Data control isn’t an enterprise upsell.
Parallel build, serial integration
Many issues build at once in isolated worktrees; they land on main one verified merge at a time. Speed where it’s safe, discipline where it counts.
Proof, not promises
This pipeline ships a real production SaaS every day. The metrics on this page are its own 90-day delivery record — not a demo.
Memory
Agents that remember your codebase — not just your last session.
Most coding agents start every task from a blank slate and re-make the same mistakes. Here, hard-won context is captured, versioned, and recalled — so the system gets sharper the longer it works on your code.
Brain
Canonical, exact-text project knowledge
Decisions, gotchas, and standing rules live in a version-controlled git repo. One hop to current truth, one hop to the reasoning behind it.
Recall bus
Cross-agent semantic memory
A self-hosted vector memory lets any agent — in any session, on any backend — recall what another already learned. Fuzzy search over hard-won context.
Signal gate
High-signal by design
The default is to write nothing. Only durable signal is captured — decisions, fixes, failed approaches — so recall stays sharp instead of drowning in transcripts.
Head to head
A different bet than the big agent platforms.
Generic agent platforms optimize for raw autonomy at enterprise scale. Otta optimizes for provable quality, persistent memory, and data you control.
Otta
Generic agent platforms
Quality gate
Otta: Enforced TDD + visual + acceptance gate — proven after the work runs
Them: “Agent-readiness” = static file-existence checks, before the work runs
Memory
Otta: Git-canonical brain + cross-agent semantic recall of decisions
Them: Filesystem persistence — no curated decision knowledge
Data control
Otta: Self-hostable on your infra, every tier
Them: On-prem is an Enterprise-only, custom-quote feature
Built for
Otta: Solo founders & 1–5 person teams shipping like a senior org
Them: Enterprise / mid-market; little for small teams
Cost
Otta: Predictable, flat plans
Them: Token metering reviewers call “a blackhole”
Tooling
Otta: Linear / GitHub / GitLab as configuration
Them: Tightly coupled to GitHub + Jira
Human checkpoints
Otta: Acceptance + screenshot evidence gate the merge
Them: Background agents; wrong assumptions surface at PR review
Comparison reflects publicly documented positioning and independent reviews of enterprise agent platforms; capabilities change over time.
Pricing Plans
Local-first by default. Hosted when you need it.
01
Free
Free
The local-first AI dev cockpit for solo builders.
Join waitlist- Native desktop app — fully local-first
- Bring your own CLI — Claude, Codex, Cursor, OpenCode
- Unlimited local background agents
- Personal DORA metrics — 30-day delivery history (own repos)
- PR cycle time + review depth — 7-day teaser
- Community support
02
Plus
$20 /mo
Your workflow follows you: synced memory, cloud runner, 90-day delivery insight — code never leaves your machine.
Join waitlistEverything in Free, and:
- Agent-native, multi-surface experience — Desktop / CLI / SDK
- One managed cloud runner — background jobs when your laptop is closed
- Memory + brain sync — cross-device, always fresh
- 90-day delivery metrics — DORA, PR cycle time, code quality trends
03
Pro
$100 /mo
Three dedicated cloud runners and full-year delivery analytics for power users and consultants.
Join waitlistEverything in Plus, and:
- Three dedicated managed cloud runners (24/7 availability)
- Full 1-year delivery history — DORA + PR review flow + code quality + Linear flow
- CI/CD deploy verification — automated health checks on every deploy
- SPACE / DevEx personal surveys — satisfaction, collaboration, flow tracking
- Early access to new features
04
Teams
Contact
For engineering leads who need delivery accountability across the whole team: shared memory, team DORA, Linear-flow dashboards, and enterprise controls.
Book a call- Multiple team members — up to 150 seats
- Custom usage limits
- Dedicated onboarding and support
- Team DORA rollups with industry benchmarks
- Linear flow metrics — cycle time per ticket, WIP, flow efficiency
- Business delivery dashboards — engineering impact mapped to OKRs
- Single Sign-On (SSO) integration
- SAML / SCIM provisioning
- Zero Data Retention (ZDR)
- Admin controls — model selection, autonomy level, access controls, deny lists
Join the Otta waitlist
Be first to run agents that test-drive, visually verify, and gate every change before it merges. We're onboarding teams in waves — early access + founder pricing.
Ship like a senior engineering org — without the org.
Join the waitlist for early access, or book a walkthrough to see the gates, the memory layer, and the live delivery record in action.