loom – LLM prompt engineering for Go

Production-ready Go library for managing, testing, optimizing, and versioning prompts for Large Language Models (LLMs). Type-safe, performant, provider-agnostic.

Install

How it works

End-to-end flow: define versioned prompts → store in Postgres or Redis → promote → run via provider → record to analytics (Postgres/Redis) → view on dashboard.

flowchart LR subgraph Define A[Prompt + Version] --> B[Store] end subgraph Registry B --> C[(Postgres)] B --> D[(Redis)] C --> E[Promote] D --> E E --> F[GetProduction] end subgraph Run F --> G[Execute] G --> H[Provider] H --> I[LLM] I --> J[Record] end subgraph Analytics J --> K[(Postgres)] J --> L[(Redis)] K --> M[Dashboard] L --> M end

Step	What happens
Define	Build prompt (ID, version, system, template, variables).
Store	Registry: Postgres, Redis, file, or memory.
Promote	Mark a version as production.
Execute	Load production prompt, call provider (OpenAI, Cerebras, etc.).
Record	Analytics: persist run (prompt_id, version, latency, success) in Postgres or Redis.
Dashboard	Charts: runs over time, success rate by prompt/version.

Step

What happens

Define

Build prompt (ID, version, system, template, variables).

Store

Registry: Postgres, Redis, file, or memory.

Promote

Mark a version as production.

Execute

Load production prompt, call provider (OpenAI, Cerebras, etc.).

Record

Analytics: persist run (prompt_id, version, latency, success) in Postgres or Redis.

Dashboard

Charts: runs over time, success rate by prompt/version.

Quick start

engine := loom.DefaultEngine() prompt := loom.New("my-prompt"). WithSystem("You are a helpful assistant."). WithTemplate("Answer: {{.question}}"). WithVariable("question", loom.String(loom.Required())). Build(engine) result, _ := prompt.Render(ctx, loom.Input{"question": "What is 2+2?"})

Registry: Postgres and Redis

// PostgreSQL db, _ := sql.Open("postgres", dsn) reg, _ := registry.NewPostgresRegistry(db, "prompts", true) // Redis rdb := redis.NewClient(&redis.Options{Addr: "localhost:6379"}) reg := registry.NewRedisRegistry(rdb, "loom:prompts") reg.Store(ctx, prompt) reg.Promote(ctx, "my-prompt", "1.2.0", registry.StageProduction) prod, _ := reg.GetProduction(ctx, "my-prompt")

Analytics: Postgres and Redis

// Persistent run history (survives restarts) // Postgres store, _ := analytics.NewPostgresStore(db, "prompt_runs") // Redis store := analytics.NewRedisStore(rdb, "loom:analytics:runs") // Server: go run ./cmd/analytics-server -store=postgres -dsn=... // Dashboard: go run ./cmd/dashboard -api=http://localhost:8080

Features

Versioning, promotion (dev → staging → production), rollback, A/B testing

Providers: OpenAI, Ollama, Anthropic, Google Gemini, Cerebras, Cohere

Registries: in-memory, file, PostgreSQL, Redis, S3

Evaluators: exact match, contains, similarity (embeddings), LLM-as-judge

Analytics: in-memory, Postgres, Redis + dashboard UI