Engineering memory for software teams
Architect AI helps your team understand how systems work, preserve why they were built, and onboard without relying on tribal knowledge.
Repository-scoped answers with source references.
Repository map
Requests pass through the JWT auth guard, which verifies the access token before workspace and role guards resolve what the caller can actually do.
The memory problem
Weeks to onboard
New engineers reconstruct the system from fragments.
Knowledge in heads
Senior engineers become the undocumented interface.
Decisions revisited
Teams lose the reasoning behind what already works.
From repository to understanding
Architect AI turns a live codebase into a searchable, explainable context layer your whole team can use.
Choose a GitHub repository.
Parse files, symbols, and relations.
Build searchable repository context.
Stream cited answers, token by token.
Streaming answers
Tokens stream live over SSE as the model responds.
Bring your own model
OpenAI, Anthropic, Grok, or Gemini—your key, your choice.
Workspaces & roles
Owner, admin, member, and viewer access per workspace.
Encrypted at rest
GitHub tokens and API keys are AES-256-GCM encrypted.
Answer feedback
Mark answers helpful or not to track quality over time.
Model-agnostic by design
We don't care who wins the AI wars.
Architect AI is model-agnostic. Bring your own provider, use hosted AI, or switch models whenever you want—your workspace, your call.
OpenAI
GPT models
Anthropic
Claude models
Grok
xAI models
Gemini
Google models
Hosted AI
No key required
Switch your active provider anytime from workspace settings—no redeploy, no lock-in.
Living onboarding guides
Every ready repository gets nine guide types—executive summary, project overview, folders, modules, services, technology stack, reading order, glossary, and pitfalls—generated automatically and regenerated on demand, all grounded in the repository and ready to ask questions from.
Generate a living guideExecutive summary
A decision-oriented overview of purpose, boundaries, core flows, and open risks— grounded in the repository and ready for source-cited follow-up questions.
Explainable by design
The goal is not to replace engineering judgment. It is to give that judgment better context—scoped to the repository or workspace you are actually working in.
Redis is used for repository indexing queues and optional retrieval caching. The indexing worker publishes jobs through BullMQ, while retrieval can cache query embeddings and assembled context.
Not another coding assistant
Copilot and Cursor answer “how do I write this faster?” Architect AI answers “how does this system work, and what happens if we change it?”
| Capability | GitHub Copilot | Cursor | Docs tools | Architect AI |
|---|---|---|---|---|
| Writes code | — | |||
| Understands architecture | ||||
| Preserves decisions | — | — | ||
| Predicts change impact | — | — | — | |
| Source-cited answers | — |
Based on our own product research; other tools evolve quickly and may close these gaps over time.
Who it's for
5–30 engineers, multiple repositories, growing technical complexity.
50–500 engineers, multiple teams, significant onboarding costs.
Self-hosted deployment, compliance controls, and advanced governance.
The product journey
Start with the questions your team is already asking. Grow toward a system that remembers the reasoning behind the software.
The question this stage should answer
How does this system work?
Simple, honest plans
Hosted AI is included on every plan. Bring your own model key to uncap AI questions and onboarding guides—the plan price stays the same.
BYOK uncaps AI questions and onboarding guides. Limits for repositories, indexing runs, workspace members, and indexing resource caps still follow the selected plan.
Frequently asked
Yes. The full stack is open source and available on GitHub, including the API, worker, indexing pipeline, and web application.
Yes. A Docker Compose stack runs the web app, API, worker, PostgreSQL, Redis, and Qdrant on your own infrastructure, with a documented single-host production deployment for AWS EC2.
Hosted AI works out of the box. Workspaces can also bring their own key for OpenAI, Anthropic, Grok, or Gemini and switch the active provider instantly—no redeploy required.
Free includes hosted AI with plan-based usage limits. PRO raises those limits for a monthly price. Enterprise is custom and starts with a conversation with sales. Connecting your own model key uncaps AI questions and onboarding guides without changing the plan price.
Repository content is parsed and embedded to power retrieval; GitHub tokens and BYOK provider keys are encrypted at rest with AES-256-GCM and are never returned in plaintext by the API.
TypeScript and JavaScript are parsed today via Tree-sitter, including TSX/JSX. Additional language parsers are on the roadmap.
Architect AI grows from repository chat into architecture exploration, decision memory, and change-impact analysis—an engineering memory layer, not just another coding assistant.
Start with what you already have
Make the system legible before the next engineer has to learn it the hard way.
Get in touch
Questions about Architect AI, partnerships, or just saying hello—send a message or find me on the web.