EXTENSIONS
Built by operatives — models, drivers, vaults, and reports, the parts that plug into Swamp.
Filter by what you need and pull what fits.
Terraform Scanner
LLM-powered Terraform code quality scanner. Analyzes repositories (local or GitLab remote) for naming convention consistency, internal module usage patterns, code organization, and Terraform best practices. Uses LiteLLM-compatible API for intelligent pattern detection and produces structured findings with severity, suggestions, and agent-consumable output.
Scaleway Inference
Manage a Scaleway Managed Inference deployment — sync state, create, update, and delete dedicated model serving deployments, and discover deployments in a region, via the Managed Inference API with X-Auth-Token auth.
Ollama
Ollama LLM integration (generate, batch generate, unload)
Anthropic
Generate text with Anthropic's Claude models via the Messages API from swamp workflows.
Meeting Summarizer
Chunk a Whisper transcript and summarize it into a Granola-style meeting note via a local Ollama LLM. Handles long recordings with greedy segment-boundary chunking and an LLM merge pass. Pure TypeScript — no Python, no subprocess. Also provides combine_notes (merge handwritten + analysis) and write_note (write markdown to disk) for full pipeline workflows.
Llm Catalog
A sourced, versioned knowledge base of LLM-ops knowledge — NOT a control plane for running models, but the structured, CEL-queryable record that running extensions read to decide which model to run where, with what settings, at what cost. Six subject kinds (model, runtime, provider, hardware, technique, endpoint), all the same uniform `entry` shape (open kind + open facets, like @stateless/inventory); a model's run-options embed on the model in runsOn[] keyed by endpoint, and capacity/plan answer "what to run where" over them. The twist: it records external, decaying knowledge, so every volatile assertion carries a provenance envelope (asOf, source, versionPins, supersededBy) — "best vLLM version" is derived as the latest un-superseded claim, never a bare rotting field. Declarative: `apply` materialises one `entry` resource per id (re-run = versioned trend); `prune` soft-retires undeclared entries.
Openrouter
Chat completions and model catalog listing against the OpenRouter LLM gateway API