Why this layer
Built into the platform, not bolted on
Every module in this layer solves the same class of problem the same way — regardless of who implements it.
01
No separate analytics pipeline
Product events, business metrics, and dashboards align to tenant data models. Skip the third-party event collector, warehouse, and BI tool chain.
02
AI with platform guardrails
LLM calls, embeddings, and prompt templates are platform modules with usage metering, versioned config, and tenant-scoped limits — not raw API keys in env files.
03
Agents that use your stack
Autonomous agents access auth, data, and tools through the same SDK as human-authored code. Execution logs and guardrails are platform defaults.
Canonical implementation
One path for every author
Junior developers, senior engineers, and AI agents produce the same standard output.
How Conflux does it
Intelligence reads from the registry, not shadow databases
Analytics events map to schema types. AI prompts version alongside application config. Agents orchestrate on runtime primitives with audit trails.
Canonical path
Event tracking aligned to platform schema definitions
Prompt templates deployed through CLI with version history
Agent tools restricted to SDK-accessible capabilities
Usage metering per tenant and application
What you get
Instrument products without a separate analytics stack
Ship AI features with metering and guardrails built in
Deploy agents that respect tenant auth and data policies
Keep intelligence close to source-of-truth data
01 · Intelligence Layer
DuckDB analytics over your SQLite
SQLite is the source of truth; DuckDB ATTACH runs aggregates for performance. Optional track events plus structured query over published tenant tables.
Analytics · One canonical path
SDK
await conflux.analytics.query(tenantId, { table, metrics: [{ fn: 'count' }] })CLI
conflux analytics stats | query --tenant … --table …
Config
# SQLite SoT; DuckDB for OLAP reads
What's possible with Analytics
01
Product and business event tracking
02
Dashboards aligned to tenant data models
03
Privacy-aware aggregation for multi-tenant apps
02 · Intelligence Layer
LLM capabilities with platform guardrails
AI MVP — bring-your-own-key. Choose Gemini or OpenAI in Console; key stored as project secret AI_API_KEY. Handlers call ctx.ai.complete; Conflux does not host models.
AI · One canonical path
SDK
await ctx.ai.complete({ prompt, system?, model? })CLI
# Console → AI, or PUT /projects/:id/ai + secrets AI_API_KEY
Config
# provider: gemini|openai|log — per project, not host env
What's possible with AI
01
BYOK: builder picks gemini|openai|log per project
02
API key in project secrets as AI_API_KEY (never host-owned)
03
ctx.ai.complete in handlers; Console GET/PUT /projects/:id/ai
03 · Intelligence Layer
Autonomous agents on platform primitives
Agents MVP — instructions + function tools; chat runs call project handlers. Console CRUD and run history via /projects/:id/agents.
Agents · One canonical path
SDK
await client.agents.chat(agentId, { message, tenantId })CLI
# use Console or management API
Config
# tools: [{ type: 'function', name }]What's possible with Agents
01
Agents with instructions and function tools
02
Chat runs invoke project handlers as tools
03
Console CRUD + run history via management API
Explore
Continue through the platform
Nine layers, one cohesive stack. Browse adjacent layers or return to the full registry.