How k-dense-ai/k-dense-byok works
K-Dense BYOK is a local-first, bring-your-own-key AI research workspace. A scientist describes a task in the web app; the backend opens a project-scoped agent session, resolves a model, and runs research work that can inspect files, execute analyses, and preserve results and notebook history. The product also includes scientific skills and workflows, evidence review, specialist subagents, compute jobs, and project cost controls. The graph combines documented capabilities with visible backend relationships; several feature implementations were not sampled, so those links are shown only when documented or left unverified.
Main entry points: session-registry.ts Live sessions · sessions.ts Session routes · run-pipeline.ts Run pipeline · models.ts Model resolution
Components 18
Research workspace
- Web workspace
web/src/app/page.tsx
Agent runtime
- HTTP API
server/src/index.ts - Session routes
server/src/api/sessions.ts - Run pipeline
server/src/agent/run-pipeline.ts - Live sessions
server/src/agent/session-registry.ts
Project and records
- Projects
server/src/projects.ts - Project files
server/src/sandbox-fs.ts - Cost ledger
server/src/cost/ledger.ts
Research capabilities
- Lab notebook
server/src/agent/notebook.ts - Evidence packages
server/src/evidence/storage.ts - Scientific skills
server/src/agent/skills.ts - Specialist agents
server/src/agent/subagents.ts
Models and compute
- Model resolution
server/src/agent/models.ts - Model providers
server/src/agent/provider-auth.ts - Compute jobs
server/src/modal/manager.ts - Compute tools
server/src/agent/modal-tool.ts
Other components
- Scientist
- Selected AI provider
Connections 19
- Scientist → Web workspace: describes task
README.md - Web workspace → HTTP API: uses
- HTTP API → Session routes: registers routes
server/src/index.ts - Session routes → Live sessions: creates sessions
server/src/api/sessions.ts - Session routes → Run pipeline: starts runs
server/src/api/sessions.ts - Run pipeline → Live sessions: pins session
server/src/agent/run-pipeline.ts - Live sessions → Model resolution: sets up runtime
server/src/agent/session-registry.ts - Model resolution → Model providers: resolves authentication
server/src/agent/models.ts - Model resolution → Selected AI provider: sends model requests
- Projects → Project files: resolves paths
server/src/sandbox-fs.ts - Run pipeline → Cost ledger: tracks run spend
server/src/agent/run-pipeline.ts - Session routes → Lab notebook: serves notebook data
- Scientist → Project files: provides research files
README.md - Run pipeline → Evidence packages: records file activity
- Live sessions → Scientific skills: activates skills
- Live sessions → Specialist agents: delegates work
- Live sessions → Compute tools: installs compute tools
server/src/agent/session-registry.ts - Compute tools → Compute jobs: submits jobs
- Compute jobs → Project files: stages project files
server/src/modal/manager.ts