01 · JUST CHATTING
Talk like to a human
No mega-prompts, no context dumps. One casual message — Product digs through your codebase and the web by itself.
Open beta · Self-hosted
Seven AI agents. One team.Already configured.
No prompt engineering: describe the task in plain words. Copula studies your code, proposes a solution and shows the result — you only validate.
- Pay in rubles
- No VPN
- Your keys and repositories
2.0 · How it works
Develop.Brainstorm.Research.Debug.Mock up.Review code.Spec it out.Test E2E.Refactor.Ship.One team for all of it.
Describe what you need in plain chat. The team plans, verifies, and delivers — you approve the plan and watch every step.
You
You — the requester
Describe the feature in plain chat. No specs, no prompt engineering — an idea is enough.
input: a message · output: intent
Product
Product — shapes the idea
Turns your message into a Business Requirement. Asks clarifying questions in chat until the scope is clear — nothing goes further without your approval.
output: BR artifact
RnD
RnD — designs the solution
Reads the BR, studies the codebase and best practices, then authors a Vision & Scope with a machine-readable jobs spec. You approve it before any code is written.
output: V&S + jobs spec
Planner
Planner — decomposes the work
A deterministic worker materializes one job per repo × iteration from the spec. No LLM guessing between design and code.
output: jobs on the kanban
Developer
Developer — writes the code
Implements each job end-to-end: edits files, runs tools, builds the project. Cheap models do the lifting, smart ones decide.
output: code + implementation notes
QA
QA — cross-verifies
Builds, runs tests, checks completion criteria. Failed job goes back to the Developer — up to 5 attempts, then escalates to you.
output: pass / redo verdict
E2E
E2E — tests the real UI
Optional stage via Chrome DevTools MCP: walks the running app like a user, validates flows unit tests can't see.
output: flow validation report
Ready
Ready — shipped
DevOps merges to your git and deploys to your servers. Every task ends with a commit, not a branch that rots.
output: merged commit + delivery
the path every feature takes — no setup, no prompt-wrangling on your side
02 · SEE IT BEFORE IT'S BUILT
Visual, not verbal
Product shows live HTML mockups and versioned artifacts right in chat — you react to what you see, not to walls of text.
03 · VALIDATE, DON'T PROMPT
You approve, they build
RnD turns the BR into a Vision & Scope with a jobs spec. Product reads it and gives you the short version — you just say yes.
Beyond shipping code
Brainstorm & spec
Talk an idea through with Product — get a BR grounded in your actual codebase.
Visual mockups
Live HTML previews iterated right in chat — see the screen before it's built.
Research & review
Point the team at any repo — it maps the codebase and reports back.
Debug & ops
DevOps can SSH into your servers, read logs, and turn findings into a fix task.
3.0 · Why Copula
A million AI agents out there.
None of them is Copula.
Everyone demos an agent that writes code. The difference is in a hundred small hacks that make an AI team actually reliable — hover any of them.
Wizard install · 6 steps Auto repo-overviews CONTEXTAuto repo-overviewsAgents read a map, not the forest. Chronicler maintains a structural overview of every repo, updated incrementally on each commit.Retry with backoff BYO keys · any provider FREEDOMBYO keys · any providerCloud credits included with every subscription — or bring your own keys: OpenRouter, custom OpenAI-compatible endpoints, local Ollama or vLLM. BYO runs at provider prices, zero markup.MCP tool servers Cross-verification QUALITY LOOPCross-verificationAgents check each other, not themselves. QA independently builds and tests what Developer wrote. Failed? Back to work. The team polices itself.Parallel jobs PARALLELISMParallel jobsOne task, many repos. Deterministic decomposition into jobs that run in parallel — no merge fights, no idle waiting.Chrome E2E stage FINAL GATEChrome E2E stageTested like a user. An agent walks the real running UI via Chrome DevTools MCP and catches what unit tests miss.Web research KNOWLEDGEWeb researchAgents google like you do. Built-in web search and page fetch — best practices, docs and error fixes flow straight into plans and code.Git auth per repo Role separation ARCHITECTURERole separation7 agents, 7 jobs. Product never touches code, DevOps is the only role allowed to mutate git. Own prompt, tools and model per role.Per-agent memory CONTINUITYPer-agent memoryNo re-onboarding from zero. Each agent keeps persistent memory across sessions — conventions, decisions, gotchas.Kanban with live jobs VISIBILITYKanban with live jobsEvery job, live. Backlog, planning, queue, in work, E2E, done — statuses stream over WebSocket as the pipeline moves. You never wonder 'is it doing anything?'Image input Mockup approvals VISUAL SIGN-OFFMockup approvalsSee it before it's built. Product renders live HTML mockups in chat — you approve pixels, not prose.Vision & Scope approval CONTROLVision & Scope approvalNo code before your OK. Every task starts from a V&S you reviewed and approved — scope never drifts silently.Artifacts, not chat logs HAND-OFFSArtifacts, not chat logsVersioned documents between agents — BR → V&S → jobs spec → QA-Notes. Each agent gets exactly its brief. Your repo stays clean of MD litter.Preflight checks Self-hosted OWNERSHIPSelf-hostedYour hardware, your rules. Code, prompts and outputs never leave your infrastructure.Smart / basic model split COSTSmart / basic model splitExpensive head, cheap hands. Smart profile for Product & RnD, basic for Developer & QA volume. Custom profiles per agent.File tree indexing Deterministic core RELIABILITYDeterministic coreLLM where creativity is needed, code where reliability is. Plan decomposition is a deterministic worker — idempotent, restart-safe, zero hallucination.Backlog → batch discuss WORKFLOWBacklog → batch discussIdeas park in the backlog. Discuss one — or merge several into a single task with one BR and one plan.Custom model profiles COSTCustom model profilesBeyond smart & basic. Create your own profiles, copy from existing ones, set per-agent overrides and custom tariffs.QA-Notes feedback loop QUALITY LOOPQA-Notes feedback loopFailures become instructions. QA writes structured notes; the Developer must read them and fix exactly what was flagged.Optimistic chat send Cost analytics TRANSPARENCYCost analyticsEvery dollar visible. Spend, tokens, iterations, lines changed — per agent, per task, per session.Polish delivered tasks REFINEMENTPolish delivered tasksDone doesn't mean frozen. One click spins a refinement task off any delivered work — source task stays intact.Change stats per commit TRANSPARENCYChange stats per commitLines added and deleted, per task. Every delivery commit is measured — churn is a first-class metric, not a mystery.Health-checked spaces MULTI-WORKSPACEHealth-checked spacesOne client, every backend. Yours, a client's, staging — health-checked before connecting, switched in one click.Session analytics drill-down TRANSPARENCYSession analytics drill-downForensics included. Open any session: requests, tokens, artifacts, per-agent cost — audit any decision the team made.
Anyone can wire an LLM to a repo.
Reliability is a hundred small hacks — we already built them.
Your repos. Your models. Your keys. Every dollar visible.
See the difference yourself →4.0 · The team
Meet the team.
Seven specialists, zero small talk.
Each role has its own prompt, its own tools, its own model profile — and its own idea of what's acceptable. They argue until it's right.
Product
Studies your project, googles like a senior — forbidden from touching code, even when it really, really wants to.
- Turns chat into BR artifacts
- Draws live HTML mockups
- Web research on best practices
RnD
Reads the whole codebase for fun. Writes the Vision & Scope, then defends every comma in it.
- Authors V&S + machine-readable jobs spec
- Grounds design in your actual code
- Architecture trade-offs, documented
Developer
The workhorse. Reads the repo map first, writes code second, complains to no one.
- Implements jobs end-to-end
- Must read QA-Notes before retrying
- Builds the project before saying 'done'
QA
The pessimist. Rebuilds everything from scratch and trusts no one's 'it works on my machine'.
- Independent build + test runs
- Writes structured QA-Notes on failure
- PASSED or NOT PASSED — no diplomacy
DevOps
The only one with git keys. Everyone else can look at the repository — through the glass.
- Sole agent allowed to mutate git
- SSH into your servers, reads logs
- Delivery itself runs as deterministic code — no LLM between QA and your repo
E2E
Clicks your UI like a very thorough user. Finds the bug you swore was impossible.
- Optional stage, per-project config
- Walks the live app via Chrome DevTools
- Spawns bug-fix tasks on failure
Chronicler
The cartographer. Maps every repo and module so the others never wander the codebase blind.
- Writes technical specs of repos & modules
- Updates them incrementally, per commit
- Purpose, conventions, pitfalls — always current
You
And what can you do? Approve plans, reject mockups, kill tasks mid-flight — and take all the credit at the demo.
- Approves the V&S before any code
- Decides when automation stalls
- Owns the keys, the repos, the bill
5.0 · Analytics
Not "87% of your token limit".
"$2.56 per shipped task".
Other tools bill you in percentages, credits and mystery limits. Copula shows the only number that matters: what a delivered task actually cost — in dollars.
Cost per task
$2.56
131 tasks · $336.80 total
Spend
$336.80
$73.60 today
Tokens in / out
36.4M / 13.2M
prompt / completion
Tasks done
131
92% success rate
Iterations
402
3.1 per task
Lines +/−
192.8k
−51.6k deleted
Cost & tokens by day
cost lines
Jul 7Jul 14Jul 21Jul 28Aug 4
Agent breakdown
AgentCostShare
Developer$152.40 45%
QA$77.60 23%
RnD$44.80 13%
Product$35.60 11%
DevOps$16.40 5%
E2E$10.00 3%
Price per outcome
Cost per task, per iteration, per repo — budget in shipped work, not in tokens.
Per-agent spend
See exactly which role burned what — and whether your model split is paying off.
Session forensics
Drill into any session: every request, token and artifact behind any decision.
Zero markup on BYO keys
Point Copula at your own provider keys and pay provider prices — the dashboard shows real cost, we add nothing on top.
6.0 · Pricing
One subscription.
Cloud credits included — or bring your own keys.
Every Copula license is a subscription with cloud credits for the built-in LLM router. Prefer your own providers? Plug in your keys on any tier and pay provider prices — zero markup. Base x1 is free while we're in beta.
Base x1
The default license. Full pipeline, free during beta.
$0 / during beta
after beta — $40/month
- 50M credits per week
- 8M credits per 5 hours
- All 7 roles, full pipeline
- Unlimited tasks & spaces
- Self-hosted or wizard cloud install
- BYO keys instead of credits — anytime
Base x5
5× the cloud quota for heavy daily shipping.
$120 / month
250M credits/week · 40M credits/5h
- Everything in Base x1
- 250M credits per week
- 40M credits per 5 hours
- BYO keys instead of credits — anytime
Base x10
The largest cloud quota for non-stop pipelines.
$200 / month
500M credits/week · 80M credits/5h
- Everything in Base x1
- 500M credits per week
- 80M credits per 5 hours
- BYO keys instead of credits — anytime
Credits are how the router meters usage: each model burns credits at its own multiplier, so a ×10 model spends 10 credits per token. Quotas run over rolling 5-hour and weekly windows and replenish continuously.
On any tier you can skip cloud credits entirely: bring your own keys — OpenRouter, Ollama, or any OpenAI-compatible endpoint — and pay your provider directly. No quotas, zero markup.
7.0 · Ship
Built for engineers who own their stack.
A team that argues until it's right. A subscription with cloud credits — or your own keys at provider prices. Hardware that's yours.