All-in-oneAI-DrivenDevelopment Platform
Compozy orchestrates your AI coding agents into one structured pipeline — from business request to merged pull request — so teams ship faster with full traceability and less rework.
Spec-Driven
DEVELOPMENT
Self-healing architecture designed for highly concurrent factory loops.
See the Full Pipeline in Action
Watch a feature go from Linear issue to merged PR — with PRD generation, parallel execution, and automated code review.
From Business Request to Merged PR.
Your team uses five AI tools with zero shared context. Compozy connects them into one auditable pipeline — so features ship with structure, not guesswork.
One Interface for Every AI Engine
Run Claude Code, Codex, Gemini, Ollama, and 8 more providers without switching tabs, re-explaining context, or losing your place.
Your team runs three AI tools in five browser tabs and none of them share context. Compozy puts 11+ providers behind one orchestration layer. Compare outputs side-by-side, swap models mid-task, and keep every interaction in a single timeline.
- 11+ providers — including local models via Ollama — in one interface
- Compare outputs side-by-side to pick the best result per task
- Swap engines mid-task without losing context or history
Issue In, Pull Request Out
Every feature follows a traceable path: Issue → PRD → Tech Spec → Tasks → PR. Nothing gets lost between handoffs.
Features don’t start with code — they start with a business request. Compozy syncs issues from Linear, runs them through a structured pipeline, and produces PRDs, Tech Specs, tasks, and a Pull Request. Every artifact links back to its source. Jira and Trello support coming soon.
- Bidirectional sync with Linear — issues, statuses, and attachments stay current
- Full audit trail from PR back through tasks, specs, and the original issue
- PRDs auto-attach to Linear so stakeholders never ask ‘where’s the spec?’
Ship Features in Parallel, Not in Sequence
Multiple agents work simultaneously — each in its own isolated git worktree — so your feature ships in one cycle, not three.
Sequential execution is the bottleneck nobody talks about. While one agent waits, everything stalls. Compozy splits features into independent tasks and runs them in parallel, each in its own worktree. One writes the API, another builds the UI, a third writes tests — all at once. Zero merge conflicts by design.
- Tasks run simultaneously across separate agents, not queued one after another
- Each agent gets its own git worktree — zero merge conflicts by design
- API, UI, and tests built at the same time by different agents
Your AI Remembers What You Already Explained
PRDs and Tech Specs act as persistent memory. Every agent starts informed — no re-prompting, no pasting the same context into every session.
The biggest hidden cost of AI-assisted development? Re-explaining your project every single session. Compozy uses structured documents as mid-term memory so agents start with full context without bloating token windows. No more copy-paste rituals. Independent executions share understanding, not token budgets.
- PRDs and Tech Specs persist across sessions as reusable project memory
- Agents start fully informed — skip the ‘let me re-explain the codebase’ tax
- Structured references replace copy-paste context dumps
OAuth2 + SSO integration with role-based access for enterprise teams. Covers IdP requirements and session management.
Architecture for provider abstraction, token refresh, and session middleware integration with Better Auth.
New analytics dashboard with real-time metrics, cost tracking, and agent performance insights.
Reusable prompt template for automated code review with security and performance checks.
The Right Model for the Right Task
Assign Opus for architecture decisions, Haiku for quick fixes, Codex for generation — at the global, issue, or individual task level.
A typo fix and an architecture decision don’t need the same model. Today’s tools force you to pick one for everything. Compozy introduces a 3-level hierarchy: set global defaults, override at the issue level, fine-tune per task. Match model power to task complexity and stop overpaying for simple work.
- 3-level hierarchy: global defaults → issue overrides → task-specific selection
- 11+ providers with no vendor lock-in — including local models via Ollama
- Cut costs by routing simple tasks to lightweight models automatically
Dedicated Agents, Not One-Size-Fits-All Chat
Each pipeline stage gets its own purpose-built agent — PRD creation, tech specs, task breakdown, and automated execution.
General-purpose AI tries to do everything and specializes in nothing. Compozy ships agents tuned for each pipeline stage. PRD Creator runs guided brainstorming. Tech Spec Creator produces architecture and risk analysis. Tasks Creator breaks specs into work units. Looper handles execution. One agent, one job, done well.
- PRD Creator, Tech Spec Creator, Tasks Creator, and Prompt Maker cover planning end-to-end
- Looper: the execution engine that runs tasks without babysitting
- Oracle (codebase Q&A) and Bug Finder (AI debugging) coming soon
New agent session
Choose an agent type to start a new session.
- PRD Creator
- Tech Spec Creator
- Tasks Creator
- Prompt Maker
- Issue Chat
- Looper
- OracleComing Soon
- Bug FinderComing Soon
AI That Thinks Before It Types
Multi-round clarification, codebase analysis, and structured approach selection — before a single line of code is written.
Most AI tools generate first and fix later. Compozy reverses that. Before producing output, the system asks clarification questions, analyzes your codebase in parallel, and presents 2–3 structured approaches with trade-offs. You pick the direction. The AI builds on a decision, not a guess.
- Multi-round clarification questions before generating anything
- 2–3 structured approaches with pros, cons, and trade-offs — you choose
- Each section validated and approved before proceeding to the next
What authentication method should the system prioritize?
Leverages existing identity providers for enterprise SSO. Stateless JWT with secure token refresh.
One Workspace for PMs and Engineers
PRDs and Tech Specs live side by side. Phase gates enforce planning before coding. Stakeholders see progress without chasing updates.
PMs write requirements in one tool. Engineers interpret them in another. Misalignment surfaces weeks later as rework. Compozy creates a shared workspace where business and engineering artifacts sit together, phase gates prevent jumping to code without a spec, and everything syncs to Linear automatically.
- Phase gates enforce the pipeline: no tasks without a spec, no spec without a PRD
- Organization-wide artifact library — browse every PRD, Tech Spec, and Prompt
- Linear sync gives stakeholders visibility without changing their workflow
OAuth2 Provider IntegrationReview Required
Architecture for multi-provider authentication with token refresh and session management.
1. Architecture Overview
Provider abstraction layer supporting Google, GitHub, and Okta SSO. Stateless JWT with refresh token rotation.
2. Security Requirements
- PKCE flow for all OAuth providers
- Token refresh with rotation policy
- Rate limiting on auth endpoints
Orchestrate anyAgentorTool.
Compozy doesn’t replace your AI coding tools. It gives them memory, context, and a shared definition of done.
AI Providers
Claude Code
Z.AI
OpenRouter
Vercel
Moonshot
Ollama
OpenCode
Codex
Gemini
Integrations
GitHub
Automatic worktree creation and repository mapping. Every agent operates in its own isolated branch — no manual setup required.
Linear
Issues sync in real-time. Status updates flow bidirectionally as work moves from planning to coding to review.
Jira
Turn Jira tickets into structured PRDs automatically. Merged PRs link back to the parent Epic for full traceability.
Confluence
Your wikis and docs become agent context. Compozy reads existing documentation before writing a single line of code.
Slack
Agents notify you only when they need a decision or a plan is ready for approval. No noise, no babysitting.
Notion
Import product specs directly. Notion pages become structured tech specs and tasks automatically.
Missing a critical part of your stack?
PredictablePricing.
Stop paying per-seat for AI editors that don’t talk to each other. Pay for one layer that orchestrates all of them.
Studio
For teams shipping products who need automated orchestration and structured workflows.
Enterprise
For organizations that need custom deployment, dedicated support, and volume pricing.
Includes Studio features +