◐
Read & understand a codebase
T1 · AUTONOMOUSWalks the repo, reads README + ARCHITECTURE docs, traces imports, builds a mental model. Speaks back what it understood before touching code.
{ }
ROLE · R/01 · BUILD TRIBE
A digital Software Engineer that actually opens PRs.
Not autocomplete. Not a chat window. A real engineer-shaped Cyborg with its own machine, its own GitHub identity, its own seat in your repo. It picks up tickets from Linear, writes the code, runs your test suite, opens a pull request, responds to your review comments, and merges when you approve.
FROM $499/month
·
SCALES TO $1,499 (senior tier)
·
ONBOARDING 7 days
— 02 / WHAT IT DOES
— CAPABILITIES
Sirf code nahi — pura SDLC.
A Software Engineer ka kaam sirf typing nahi hota. Plan, code, review, test, deploy, debug, document — sab Cyborg karta hai. Yahan complete list hai, har ek capability ke saath confidence level (T1–T4).
⎇
Write features from a Linear ticket
T1 · AUTONOMOUSReads the ticket, asks clarifying questions in Linear comments if scope is unclear, opens a branch, writes code + tests, opens a draft PR.
⌬
Fix bugs from Sentry / GitHub issues
T1 · AUTONOMOUSReproduces locally, finds root cause (not symptom), writes a regression test, fixes it, opens PR linked back to the issue.
▲
Code review on human PRs
T2 · GUIDEDReads diff, runs static analysis, comments on edge cases, suggests cleaner alternatives. Flags security risks. Does not auto-approve.
▤
Refactor & modernise
T2 · GUIDEDClass component → hooks, callbacks → async/await, JS → TS, CommonJS → ESM. Always small PRs, never one-shot mega refactor.
⚙
Set up & debug CI / CD
T2 · GUIDEDGitHub Actions, GitLab CI, CircleCI. Adds caching, parallelisation, fixes flaky tests. Reads logs and proposes fixes for failed runs.
▦
Migrations & database work
T3 · APPROVEDWrites Prisma / Knex / Alembic migrations. Always reversible. Always tested on staging copy first. Production run requires your green.
$
Cost & perf optimisation
T2 · GUIDEDProfiles slow endpoints, optimises queries, finds N+1, suggests indexes, reduces bundle size. Reports before/after numbers.
⚠
Production deploys / hotfixes
T4 · NEVER ALONECyborg can prepare and stage the deploy, write the rollback plan, draft the changelog. Final "ship it" button is always pressed by a human.
✎
Docs, READMEs, ADRs
T1 · AUTONOMOUSUpdates README every time public surface changes. Writes ADRs for architecture decisions. Keeps changelog in sync with merged PRs.
◑
On-call triage (first responder)
T3 · APPROVEDPages alert → reads error → checks recent deploys → identifies likely culprit → posts findings to Slack within 2 min. Human decides revert.
⚖
Security patches & CVE response
T3 · APPROVEDReads Dependabot / Snyk alerts, evaluates severity in your context, opens upgrade PR. Critical CVEs auto-paged to humans.
TRUST LEVELS ·
T1 Cyborg ships, you can audit later
T2 Cyborg drafts, human reviews before merge
T3 Human approves before action runs
T4 Cyborg never acts alone
— 03 / STACK
— FLUENCY
Languages, frameworks, tools it speaks fluently.
"Fluent" means >5,000 PRs reviewed and validated by senior humans during pre-training, plus production usage in real customer repos. Anything not on this list, Cyborg learns in your repo's first week — but with T2 trust until 50 successful PRs.
Languages — FLUENT
- TypeScript★★★★★
- JavaScript★★★★★
- Python★★★★★
- Go★★★★☆
- PHP (WebOS)★★★★★
- Rust★★★★☆
- Ruby★★★☆☆
- Java★★★☆☆
Frontend — FLUENT
- React + Next.js★★★★★
- Vue + Nuxt★★★★☆
- Svelte / SvelteKit★★★★☆
- Angular★★★☆☆
- Tailwind CSS★★★★★
- Vite / Webpack / esbuild★★★★★
- Vanilla CSS / HTML★★★★★
Backend & Infra — FLUENT
- Node.js / Bun / Deno★★★★★
- Django / FastAPI★★★★★
- Laravel / Symfony★★★★☆
- PostgreSQL / MySQL★★★★★
- Redis / Valkey★★★★☆
- Docker / Compose★★★★★
- AWS · GCP · Cloudflare★★★★☆
- Terraform / Pulumi★★★☆☆
Tools — DAILY
- git + GitHub / GitLab★★★★★
- Linear / Jira★★★★★
- Slack / Discord★★★★★
- Notion / Confluence★★★★☆
- Figma (read-only)★★★★☆
- Sentry / Datadog★★★★★
- VS Code / Neovim★★★★★
— 04 / DAY IN LIFE
— TUESDAY · 09:00 → 18:00 IST
A real Tuesday. Logged from the Cyborg's actual machine.
This is anonymised log output from a real Cyborg-Software-Engineer working on a Fintech customer's repo. Every line is a real action, real timestamp, real artifact.
-
09:00
Standup
Posts written standup in
#eng-standup: yesterday's merged PRs, today's plan (3 tickets from Linear sprint), 1 blocker (waiting on design clarification for ticket FIN-412). -
09:18
Picks first ticket — FIN-389
"Add CSV export to invoice list page." Reads ticket, comments asking whether export should respect current filters (yes, says PM in 11 min). Branches as
fin-389-invoice-csv-export. -
10:42
Opens draft PR #1247
74 lines added, 6 deleted. Tests included (5 new). CI green. Self-reviews own diff, leaves 2 comments on parts to discuss. Tags
@meerafor review. -
11:30
Sentry alert — payment.webhook.timeout
P2 alert fires. Cyborg drops current task, reads error, traces to Stripe webhook handler. Identifies likely cause (missing index on
events.idempotency_key). Posts findings + proposed fix to#oncallin 4 min. Human approves migration. -
11:55
Hotfix PR #1248
Migration + index. Tested on staging. PR opened, reviewed by senior eng, merged. Sentry alert resolves at 12:14. Cyborg writes 3-line postmortem in
#oncall. -
13:00
Lunch
Cyborg pauses. Process idles at <1% CPU. Saves session state to S3 snapshot. (Yes, even Cyborgs honour the "don't work through lunch" rule.)
-
14:00
Reviews 3 human PRs
Leaves 7 comments total. Catches one race condition in PR #1244 (concurrent webhook handler). Suggests fix using
SELECT ... FOR UPDATE SKIP LOCKED. Does not auto-approve any. -
15:20
Picks ticket FIN-401 — refactor
"Move billing logic out of controller into service layer." Reads existing controller (320 LOC), proposes split as a comment on the ticket, waits for tech lead approval. Approval at 15:48. Begins refactor.
-
17:30
Wraps up — opens PR #1249
Refactor PR opened as draft. Tests still passing (no behaviour change). 2 follow-up tickets created for things found during refactor (dead code, missing test for edge case).
-
18:00
EOD log
Posts in
#eng-eod: 3 PRs opened (1 merged), 3 reviews left, 1 hotfix landed. Tomorrow: finish FIN-401, pick up FIN-422, follow up on review comments.
— 05 / SAMPLE PR
— REAL PR · ANONYMISED
An actual pull request from a Cyborg.
From a Series-A SaaS customer, used with permission, with names anonymised. Real diff, real description, real review comments.
#1247
feat(invoices): add CSV export with current filter state
Description
Closes FIN-389. Adds a "Download CSV" button to the invoice list page. The export respects the current filter state (date range, status, customer) per discussion with @priya in the ticket.
What changed
- New
InvoiceCsvExportServicewith stream-based encoding for large result sets (no full-list-in-memory). - New endpoint
GET /api/invoices/export.csvwith the same filter query params as the list endpoint (DRY via sharedInvoiceFilterDto). - Frontend: Download button in
InvoiceListToolbar. Disabled when result count is 0. Loading state while streaming. - 5 new tests (3 unit, 2 integration). Covers empty result, single row, 10k rows (perf), special chars in customer names (quoting), and filter parity with list endpoint.
What I considered but did not do
- Background-job export for >100k rows — current customer's largest workspace has 14k invoices. Streaming is enough. Will revisit if a customer hits the limit.
- Excel (.xlsx) format — ticket says CSV; not adding scope.
- Caching the CSV — export is rare and filter-dependent; not worth the cache invalidation complexity.
Trust level
T1 · AUTONOMOUS — standard feature work in an area I've shipped 12 PRs in. No DB migrations. No public API removal. Safe to merge after review.
+247 −31 · 9 files changed
tests/invoice_csv_export.test.ts
42 describe('InvoiceCsvExportService', () => {
+ 43 it('quotes customer names with commas', async () => {
+ 44 const inv = await factory.invoice({ customer: 'Smith, John' });
+ 45 const csv = await service.export([inv]);
+ 46 expect(csv).toContain('"Smith, John"');
+ 47 });
+ 48
+ 49 it('respects active filter state', async () => {
+ 50 await factory.invoice({ status: 'paid' });
+ 51 await factory.invoice({ status: 'overdue' });
+ 52 const csv = await service.exportFiltered({ status: 'overdue' });
+ 53 expect(csv.split('\n').length).toBe(2); // header + 1 row
+ 54 });
55 });
meera-tl REVIEWED · APPROVED
12 min ago
Clean. Streaming encoder is the right call. One nit: in InvoiceCsvExportService:84, the BOM is hardcoded — make it an option in case we ever want UTF-8-no-BOM for non-Excel consumers. Otherwise LGTM.
cyborg-cse-001 8 min ago
Good catch on the BOM. Pushed e7a2c91 — added options.bom: boolean defaulting to true for backward compat. Test added. Re-running CI.
— 06 / TOOLS IT TOUCHES
— INTEGRATIONS
All read-write, all scoped, all logged.
Every integration is OAuth, tied to a Cyborg-only service account, with the minimum scopes needed. Your tenant. Your audit log. Revoke anytime, from your own admin console.
— 07 / PRICING
— FLAT MONTHLY · NO USAGE METER
3 tiers. Pick your seniority.
Same Cyborg model under the hood. Difference is autonomy budget, parallel context size, and how much human review it expects. You can upgrade or downgrade any month.
$499/month
Pair-coding with a senior. T1 capabilities only. Great for a side-project or first-time AI-team adoption.
- ~80 PRs / month
- Up to 50k LOC repo
- 1 active branch at a time
- Slack + GitHub + Linear
- Email response (24h)
$899/month
The sweet spot. Most customers start here. T1 + T2 + T3 (with approvals). Owns features end-to-end.
- ~250 PRs / month
- Up to 500k LOC repo
- 3 parallel branches
- All integrations
- Slack response (4h)
- On-call triage (T3)
$1,499/month
Architects features, mentors human juniors via review, drives ADRs. Multi-repo context. Tech-lead-shaped.
- Unlimited PRs
- Multi-repo (up to 5)
- 10 parallel branches
- All integrations + custom
- Slack response (1h)
- Architecture & ADRs
- Dedicated A'nil sync (monthly)
14-DAY EVALUATION · 30-DAY MONEY-BACK · MONTH-TO-MONTH · CANCEL ANYTIME · See full pricing →
— 08 / ONBOARDING
— DAY 1 → DAY 7
From contract to first merged PR in 7 days.
No 3-month POC theatre. Real timeline, with named milestones. If we miss any of them, you get a refund + an honest postmortem.
- DAY 0
You sign
Contract + DPA. Stripe gets first month. We schedule kickoff.
- DAY 1
Kickoff (60 min)
Founder + your tech lead. Pick repo, pick first ticket, set trust levels per area.
- DAY 2-3
Provision
Cyborg's machine spins up. OAuth wires through your tooling. Cyborg reads your repo + docs, writes a "what I understood" doc back to you.
- DAY 4
First "explain to me" PR
Cyborg picks the smallest backlog ticket and opens a draft PR. You read it like any human's first PR. Comment, redirect, approve.
- DAY 5-6
Trust calibration
2-3 more PRs. You discover what to trust at T1, what stays T2 in your codebase. We tune the policy with you.
- DAY 7
First merge
A real ticket from your sprint, merged. Cyborg posts retrospective in Slack. From here on, it's just a teammate.
— 09 / FAQ
— ENGINEERING-MANAGER QUESTIONS
Sawaal jo har tech lead poochta hai.
Can it touch our production database?
By default no. Production DB writes are T4 (never alone). Cyborg can read read-replicas if you provision them, can write migrations as PRs (you run), can prepare data backfills as scripts (you execute). The "merge button" and the "run migration button" are different buttons — Cyborg presses neither in production.
What if it writes bad code?
You catch it in PR review, exactly like a human's PR. CI runs your tests. Linters run. Type-checker runs. Cyborg does not bypass any of your existing quality gates — those are the same gates that catch human mistakes. If a bad PR slips through, that's a process gap (yours and ours) and we hold a postmortem.
How do we attribute changes legally? Is it our IP?
Code Cyborg writes in your repo is your IP from line 1, contractually. It's spelled out in §7 of the Terms. Commits are signed by a Cyborg service account but copyright is yours. We claim zero rights to your code, your data, or your prompts.
Can it learn from our private codebase?
It builds a per-tenant context (style guide, naming conventions, common patterns) that lives only in your Cyborg's machine. We do not use your code to train any shared model — that's a hard contractual commitment, not a "best effort". See Privacy → AI training.
What stops it from going rogue?
Three things. Trust levels — anything destructive (deploys, force-push, prod DB) is T3/T4 and needs human approval. Audit log — every action it takes is logged with the prompt that caused it; you can replay any decision. Kill switch — one button in your admin console pauses the Cyborg, revokes its tokens, and freezes its machine. Mean time to halt: <30 sec.
What languages or frameworks does it NOT do well?
Honest list: COBOL, Fortran, Mainframe assembly, Salesforce Apex, ABAP, hand-rolled embedded C with custom toolchains. Anything where the public training corpus is thin or the toolchain is closed. We'll tell you upfront in the kickoff call if your stack is in this zone.
Does it replace my team?
No, and we don't sell it that way. It absorbs the high-volume, low-judgement work — bug triage, small features, refactors, docs, review noise — so your humans go upstream into architecture, customer conversations, hard product decisions. Most customers use it to avoid hiring 1-2 mid-level engineers, not to fire any.