Multica — Tech · Product Feature Research

Architecture

┌──────────────┐     ┌──────────────┐     ┌──────────────────┐
│   Next.js    │────>│  Go Backend  │────>│   PostgreSQL     │
│   Frontend   │<────│  (Chi + WS)  │<────│   (pgvector)     │
└──────────────┘     └──────┬───────┘     └──────────────────┘
                            │
                     ┌──────┴───────┐
                     │ Agent Daemon │  runs on YOUR machine
                     └──────────────┘  (Claude Code, Codex, OpenCode, …)

Source: GitHub README, captured 2026-04-28.

Stack

Layer	Tech
Frontend	Next.js 16 (App Router)
Backend	Go (Chi router, sqlc, gorilla/websocket)
Database	PostgreSQL 17 with pgvector
Agent runtime	Local daemon executing user-installed CLI
Streaming	WebSocket (gorilla/websocket)

Supported agent CLIs (built-in providers)

Eleven listed in docs, ten in the README — minor doc drift. Authoritative list per Welcome:

Claude Code · Codex · Cursor · Copilot · Gemini · Hermes · Kimi · Kiro CLI · OpenCode · OpenClaw · Pi

The daemon auto-detects whichever of these are on PATH. Each is a "provider" the agent can be configured against.

Deployment shapes

Shape	What	When
Multica Cloud	Managed backend, user installs CLI + daemon locally, connects to multica.ai server	Waitlist as of 2026-04-28
Self-host (Docker)	Docker Compose, full backend on user's infra	Documented + recommended for regulated teams
Self-host (Kubernetes / Helm)	Full backend on user's k8s	Mentioned, doc URL not captured
Desktop app	Native multi-tab window, CLI built-in, daemon starts on launch	"Recommended" per docs

❓

license. The README doesn't state it inline; the GitHub repo would have a LICENSE file. Apache 2.0 or MIT are the standard guesses for OSS-first projects of this shape.

Developer surface

CLI: multica login · multica daemon start · multica daemon status · multica setup · multica setup self-host · multica issue list · multica issue create · multica update.
Agent daemon: runs locally, drives user-installed coding-agent CLIs.
HTTP API: Go backend with Chi router; websocket endpoint for real-time event streaming.
Skills: declared in SKILL.md files with Steps sections; agents reference and execute them.

What's NOT advertised

No mention of LangChain, OpenAI Agents SDK, MCP, or x402 integration in captured pages. However, the GitHub navigation surfaces "MCP Registry" — possibly an MCP integration exists or is in flight.
No mobile app (web-only on the user side; native desktop is the alternate client).
No GPU / inference provisioning (they don't host inference; user pays their own LLM provider).

❓

whether Multica integrates with MCP servers, and whether agents can call HTTP endpoints with auth (relevant for HTTP 402 / MPP integration).

Strategic note for CreateOS

💭

Multica's daemon-on-user-machine architecture means CreateOS could integrate as a "cloud runtime" — a Multica-compatible runtime endpoint that runs agents in CreateOS sandboxes, settles via MPP, and reports task progress over WebSocket back to Multica. This is a clean partnership shape and a Tier 2 BD candidate (see CreateOS - BD Targets).

Sources

github.com/multica-ai/multica
multica.ai/docs (Welcome)