Sema packages

First-party packages for Sema — a Lisp with first-class LLM primitives.

First-party packages for Sema, published to the registry under the official sema account. This is a monorepo: each sema-*/ directory is an independent package (its own sema.toml, version, and tests) with no cross-dependencies. Install any of them with:

sema pkg add sema-<name>

Packages

AI / LLM integrations

Service wrappers for the things LLM and agent builders reach for — RAG, memory, tools, media. All are thin REST clients over Sema's HTTP + JSON stdlib.

Package	Description
sema-firecrawl	Scrape, crawl, map, and search the web through the Firecrawl API
sema-tavily	Web search and content extraction tuned for LLMs (Tavily)
sema-exa	Neural/semantic web search and page contents (Exa)
sema-qdrant	Vector store: collections, upsert, filtered search (Qdrant REST)
sema-upstash-vector	Connectionless HTTP-native serverless vector store (Upstash)
sema-mem0	Long-term memory for agents: add, search, manage (Mem0)
sema-replicate	Run hosted models: create predictions, poll, fetch outputs (Replicate)
sema-elevenlabs	Text-to-speech + voices; returns audio bytes (ElevenLabs)
sema-unstructured	Documents → structured elements for RAG (Unstructured.io)
sema-graphql	Minimal generic GraphQL client (queries + mutations over HTTP)
sema-cdp	Chrome DevTools Protocol client: launch, connect, send commands, wait for events
sema-browser	Drive a real browser over CDP: navigate, click, fill, screenshot, agent-facing snapshots

Embeddings, reranking, and classification are built into Sema (llm/embed, llm/rerank, llm/classify — Cohere/Jina/Voyage/OpenAI providers), so there's no package for those.

Utilities & tooling

Package	Description
sema-policies	Reusable least-privilege, content-safety, output, and workflow-evidence policies
sema-medley	Utility belt: map/list transformations the stdlib doesn't cover
sema-strings	Case conversion, slugs, padding, wrapping, string distance
sema-test	Test framework — define, run, and report tests
sema-validate	Schema validation for maps and data structures
sema-config	Layered config: files, env vars, profiles, reference expansion
sema-dotenv	Parse .env files and load environment variables
sema-log	Leveled, structured logging with pluggable appenders
sema-retry	Retry with exponential backoff, circuit breaker, rate limiting
sema-cache	TTL/LRU caching, memoization, optional persistent kv stores
sema-cli	CLI argument parsing: flags, options, subcommands
sema-url	URL parsing, building, resolving, query-string manipulation
sema-html	S-expression → HTML generation, Hiccup-style
sema-fmt	Pretty-printing, table formatting, tree rendering
sema-frida	Bounded Frida inventory and native persistent observation

Service clients

Package	Description
sema-github	GitHub REST API v3 client with pagination and typed errors
sema-asana	Asana REST API (v1): tasks, projects, workspaces, comments

Developing

Conventions every package follows are in STYLE.md. To start a new package, copy package-template/ and run its scaffold.sema (see SCAFFOLD.md).

./test-all.sh          # link every package into the cache and run all test suites
./publish-all.sh       # publish each package at its current version (after login)

test-all.sh copies the local packages into ~/.sema/packages/ first (the resolver rejects symlinks), so short-name imports like (import "sema-test") resolve during development.

Publishing

Packages are published to the registry under the sema account (which grants the "Official" badge). Bump version in a package's sema.toml and add a CHANGELOG.md entry for each release, then:

sema pkg login --token <token>    # token for the sema account
./publish-all.sh                  # immutable versions are skipped

License

Every package is MIT.