Arcana is live · free · last verified 24 Oct 2026 · viability monitor
Arcana
Embeddable RAG for Elixir and Phoenix — vector search, GraphRAG, and agentic retrieval inside your existing Ecto/Postgres repo.
If your stack is Elixir, Phoenix, and Postgres, Arcana is the least-friction way to get real RAG into production — no second database, no second language, no separate service to page. The three surfaces cover a lot of ground: search/ask for the common case, Pipeline when you need step-level control (rewrite → select → decompose → search → rerank → answer → ground), and Loop when the query itself is open-ended. GraphRAG, local Bumblebee embeddings, and built-in NLI grounding scoring are unusual at this layer. Non-Elixir teams get nothing here; they should reach for a Python framework.
Verdict generatedVerified 6d ago · liveness 64/100 · cite: rightaichoice.com/tools/arcana
- Elixir/Phoenix teams already running Ecto and PostgreSQL who want RAG without new infrastructure
- Teams that need vector search, GraphRAG, and agentic retrieval inside an OTP supervision tree
- Developers who want local-first embeddings and reranking with an optional cloud LLM swap
- Apps where open-ended or multi-hop questions need an LLM to pick its own search sequence
- Non-Elixir projects — a Python RAG framework or a hosted retrieval service fits better
- Stacks built on non-Postgres vector stores
- Elixir newcomers not ready to own supervision trees and behaviour wiring
We scan live Reddit threads, YouTube comments, X posts, G2 reviews and other communities — and hand you an honest verdict in under a minute.
- Honest verdict, not marketing
- Real pros & cons from real users
- Attributed quotes with receipts
3 free scans · no card needed
Skip Arcana if your app isn't Elixir or Phoenix, or if you've standardized on a vector database other than Postgres with pgvector — there is no path in from either starting point.
Running local Bumblebee embeddings and the cross-encoder reranker means paying for the CPU or GPU, memory, and node capacity that the Nx.Serving processes consume in your own infrastructure.
Arcana is an open-source library, so the cost is your Postgres instance plus the compute for local embeddings and the provider bill for any cloud LLM you address through model strings. For an Elixir team already paying for both, that is typically far less than a managed retrieval service with its own infrastructure line item.
In short
Arcana — Embeddable RAG for Elixir and Phoenix — vector search, GraphRAG, and agentic retrieval inside your existing Ecto/Postgres repo. Best for Elixir/Phoenix teams already running Ecto and PostgreSQL who want RAG without new infrastructure, Teams that need vector search, GraphRAG, and agentic retrieval inside an OTP supervision tree, Developers who want local-first embeddings and reranking with an optional cloud LLM swap. Free to use.
What is Arcana?
Arcana — Embeddable RAG for Elixir and Phoenix — vector search, GraphRAG, and agentic retrieval inside your existing Ecto/Postgres repo. Best for Elixir/Phoenix teams already running Ecto and PostgreSQL who want RAG without new infrastructure, Teams that need vector search, GraphRAG, and agentic retrieval inside an OTP supervision tree, Developers who want local-first embeddings and reranking with an optional cloud LLM swap. Free to use.
Arcana is an open-source retrieval-augmented generation library built natively for Elixir and Phoenix. It adds vector search, knowledge graphs, and LLM-driven retrieval to any app that already has an Ecto repo — no separate vector database, no external indexing service, no Python sidecar glued on over HTTP. Documents, chunks, embeddings, and the knowledge graph live as tables in your existing Postgres database with pgvector. The library ships three usage shapes mapped to published RAG patterns. Arcana.search/2 and Arcana.ask/2 are the default door: one call with query rewriting, hybrid semantic-plus-fulltext search, graph fusion, and cross-encoder reranking. Arcana.Pipeline composes your own steps (gate, rewrite, select, expand, decompose, search, reason, rerank, answer, ground) with each step a replaceable behaviour. Arcana.Loop hands control flow to an LLM that picks tools — search, answer, give_up — each turn until it commits, best for open-ended or multi-hop questions where the right search order isn't obvious upfront. GraphRAG is included: entity extraction, relationship linking, and Leiden community detection, with graph results fused into vector search. Embeddings and the cross-encoder reranker run local-first through Bumblebee on EXLA, EMLX (Apple Silicon), or Torchx; cloud LLMs are addressed through model strings such as openai:gpt-4o-mini and zai:glm-4.5-flash. Embedders and rerankers are Nx.Serving instances under your OTP supervision tree, and every step emits :telemetry spans you can reattach to LiveDashboard. This is for Elixir and Phoenix teams who want AI search and Q&A inside an app they already operate, and who would rather supervise one more process than run one more service.
Key features
- Advanced RAG via Arcana.search/2 and Arcana.ask/2 with query rewriting
- Hybrid semantic plus fulltext retrieval fused with Reciprocal Rank Fusion
- Cross-encoder reranking via Bumblebee, local by default
- GraphRAG with entity extraction, relationship linking, and Leiden community detection
- Graph fusion that merges knowledge-graph results into vector search
- Modular RAG via Arcana.Pipeline with replaceable step behaviours
- Pipeline steps: gate, rewrite, select, expand, decompose, search, reason, rerank, answer, ground
- Agentic RAG via Arcana.Loop with runtime tool selection, tool_history, and terminated_by
- Controller/answerer split — cheap fast controller model plus a stronger answering model
- NLI hallucination check via Pipeline.ground and Loop.ground with 0.0-1.0 faithfulness score
- Self-correction modes (CRAG-inspired) on Pipeline.search/2 and Pipeline.answer/2
- Local embeddings via Bumblebee on EXLA, EMLX (Apple Silicon), or Torchx
- Cloud LLM model strings such as openai:gpt-4o-mini and zai:glm-4.5-flash
- Pluggable behaviours for chunker, embedder, searcher, reranker, and answerer
- Ingestion background jobs with Oban and an Igniter installer for migrations and dashboard route
Behind the verdict
Arcana's argument is architectural, and it mostly holds up. Because documents, chunks, embeddings, and the knowledge graph are just tables in the Postgres database your Phoenix app already runs, there is no second datastore to back up, migrate, or page someone about. pgvector does the ANN work. Embedders and rerankers are Nx.Serving instances sitting in the supervision tree you already own, and every retrieval step emits :telemetry spans that drop straight into LiveDashboard — the same observability story as the rest of the app. What you actually get is three doors into the same pipeline, mapped explicitly onto the Singh et al. 2025 Agentic RAG survey slots. Arcana.search/2 and Arcana.ask/2 are Advanced RAG with defaults already applied: query rewriting, hybrid semantic plus full-text retrieval fused with Reciprocal Rank Fusion, optional graph fusion, and cross-encoder reranking. Arcana.Pipeline is Modular RAG — named steps (gate, rewrite, select, expand, decompose, search, reason, rerank, answer, ground) you compose with a pipe operator, each backed by a single-callback behaviour you can swap. Arcana.Loop is Agentic RAG: a controller LLM picks among search, answer, and give_up each turn, returning tool_history and a terminated_by value of :answered, :gave_up, :max_iterations, or :error. The router/answerer split is supported, so a cheap controller model can pick tools while a stronger model writes the reply. Two details are worth calling out. First, Pipeline.ground and Loop.ground run an NLI faithfulness check and return a 0.0–1.0 grounding score, which is the closest thing here to an automatic hallucination guard — you can gate on it before showing an answer. Second, GraphRAG is not a bolt-on: entity extraction, relationship linking, and Leiden community detection happen over the same corpus, and graph results are fused back into vector search. Where it bites is scope. Everything is Elixir, everything assumes Postgres with pgvector, and a stack standardized on another vector database has no path in. The agentic mode is only as good as the controller you point at it — the documentation itself notes that a cheap fast controller makes weaker tool choices than a stronger model. And local-first means compute and memory planning land on your app rather than a vendor's, so you need someone comfortable owning supervision trees and behaviour wiring. There is deliberately no Naive RAG one-shot mode; even the simplest call rewrites and reranks. If you wanted dumb embed-retrieve-generate, this is not that.
Models under the hood
as of 2026-09-08
Limitations
- Arcana is tied to Elixir and Phoenix, so non-Elixir projects cannot use it at all.
- It assumes PostgreSQL with pgvector as the store for documents, chunks, embeddings, and the knowledge graph; projects standardized on another vector database have no path here.
- Agentic quality depends on the controller model you configure — the docs note that cheap controllers produce weaker tool choices than stronger ones.
- The library deliberately ships no Naive RAG one-shot mode, so even the simplest entry point rewrites, reranks, and fuses.
- Local embeddings and the cross-encoder reranker mean running Nx.Serving processes under your supervision tree, which adds compute and memory planning to your app rather than offloading it to a vendor.
as of 2026-10-13
Researching Arcana? Get your full AI stack in 60 seconds.
Free, no signup — tell us your goal and get tools matched to your budget & existing stack.
Is Arcana still active?
Arcana is live · free · last verified 24 Oct 2026 · viability monitor.
Viability score
How well maintained and how widely used is Arcana? Built from what the vendor actually publishes (docs, changelog, tutorials, integrations, pricing), whether the site is live, and how much real users discuss it. How we calculate this
Last calculated: October 2026
How we score →What do users say about Arcana?
What people actually say about Arcana — is it worth it?
Our community sentiment research written Jul 3, 2026 is more than 90 days old, so it is being re-reviewed against current data and is not shown.
How we verified this
Every listed fact about Arcana— pricing, features, links, status — is re-read from the vendor's own pages by our verification pipeline and each change is logged with a date; the last pass ran on 24 Oct 2026. Pages mix model-assisted synthesis with human review; see how we verify.
Verification history
We have re-verified Arcana 11 times since . Each pass re-reads the vendor's own pages and re-checks every listed field against that evidence; passes where nothing had changed are marked as such.
- — changed models, switching out, integrations, summary, switching in, limitations, setup time, pricing fit, pricing tiers, our verdict, who should skip it, who it suits, features, description, hidden costs
- — changed who should skip it, who it suits, description, our verdict, features, integrations, pricing tiers, summary
- — changed domain expiry, SSL certificate
- — changed integrations, switching out, pricing tiers, pricing fit, description, setup time, who should skip it, who it suits, use cases, limitations, switching in, hidden costs, models, features, our verdict, summary
- — changed domain expiry
- — changed link health
Showing the 6 most recent of 11 verification passes.
Free to cite with attribution — this page re-verifies continuously.
How much does Arcana cost in 2026?
Arcana: Free to use. Pricing last verified 24 Oct 2026.
Plans compared
Every published Arcana tier, its price as the vendor publishes it, who it fits, and what it adds over the tier below. The 12-month cost projection follows.
| Plan | Price | Billing | Who it's for |
|---|---|---|---|
| Open Source | Free | — | Elixir and Phoenix teams running Ecto and Postgres who want retrieval inside their own app without a second datastore or a paid retrieval service.Adds: Starting tier — the full library at no cost, including all three RAG surfaces, GraphRAG, local Bumblebee embeddings, cloud LLM model strings, and the LiveView dashboard. |
12-month cost
Project the real annual outlay, including the implied monthly cost when only an annual tier is published.
Vendor list price only. Add-on usage, seat overages, and contract minimums are surfaced under Hidden costs & gotchas.
Where the pricing makes sense
The company stage and team size where Arcana's pricing actually pencils out — and where peers do it cheaper.
Arcana is an open-source library, so the cost is your Postgres instance plus the compute for local embeddings and the provider bill for any cloud LLM you address through model strings. For an Elixir team already paying for both, that is typically far less than a managed retrieval service with its own infrastructure line item.
Has Arcana's pricing changed?
No plan price for Arcana has moved since we began logging in Aug 2026; the pricing page is re-read on every verification pass.
Who should pick Arcana?
Pick Arcana if you are Elixir/Phoenix teams already running Ecto and PostgreSQL who want RAG without new infrastructure, Teams that need vector search, GraphRAG, and agentic retrieval inside an OTP supervision tree, Developers who want local-first embeddings and reranking with an optional cloud LLM swap, Apps where open-ended or multi-hop questions need an LLM to pick its own search sequence.
Look elsewhere if you are Non-Elixir projects — a Python RAG framework or a hosted retrieval service fits better, Stacks built on non-Postgres vector stores, Elixir newcomers not ready to own supervision trees and behaviour wiring, Projects wanting a Naive RAG one-shot mode — Arcana always adds rewriting, reranking, and fusion.
Skip it if Skip Arcana if your app isn't Elixir or Phoenix, or if you've standardized on a vector database other than Postgres with pgvector — there is no path in from either starting point.
GeneratedReal-world workflow fit
Concrete scenarios for the personas Arcana actually fits — and what changes day-one when you adopt it.
You add Arcana to an existing app, run the Igniter installer for migrations and the dashboard route, then ingest docs and error logs into a pgvector-backed collection before wiring Arcana.ask/2 into an endpoint.
Outcome: Source-cited answers over your own corpus, with embeddings staying in the Postgres database your app already operates and telemetry visible in LiveDashboard.
You drop from Arcana.ask/2 into an Arcana.Pipeline that gates, rewrites, decomposes a comparison question into sub-questions, searches, reranks with the cross-encoder, answers, and finishes with Pipeline.ground for an NLI faithfulness score.
Outcome: Step-level control over order and behaviour, plus a 0.0-1.0 grounding score you can gate on before showing the answer.
You run Arcana.Loop.new/2 across several collections and let a controller model pick among search, answer, and give_up each turn while a stronger model writes the final reply.
Outcome: tool_history shows exactly which searches the model chose, terminated_by tells you whether it answered, gave up, or hit max_iterations, and you can score faithfulness with Loop.ground.
Use cases
- Add source-cited Q&A over your Phoenix app's own documentation and error logs
- Semantic search for an Elixir forum, knowledge base, or internal wiki using pgvector tables you already own
- Multi-hop research questions where Arcana.Loop decides which searches to run before answering
- Customer support reply drafting grounded with an NLI check so unsupported claims are caught
- GraphRAG over a document corpus where entity relationships matter alongside vector similarity
- Self-hosted internal research assistant where embeddings never leave your infrastructure
- Wire custom retrieval steps into an existing OTP app with single-callback behaviours
- Debug and evaluate retrieval quality through the LiveView dashboard and telemetry spans
Setup time & first value
How long it actually takes to get something useful out of Arcana — broken out by persona, not the marketing-page minute.
A Phoenix developer already running Ecto and Postgres can typically get from install to a first ingested document and an Arcana.ask/2 answer in an afternoon — the Igniter installer handles migrations and the dashboard route, and local embeddings run under your existing supervision tree. Custom Pipeline steps and graph tuning take longer.
Switching to or from Arcana
How to bring data in from common predecessors and how to get it back out — written for the switcher, not the buyer.
- →From naive embed-retrieve-generate code in your own Phoenix app: replace the hand-rolled retrieval path with Arcana.search/2 or Arcana.ask/2, which adds rewriting, hybrid search, and reranking on top of the same
- →From chunk and embedding rows you already keep in Postgres: move them into Arcana's document, chunk, and embedding tables so pgvector search and the knowledge graph read from one schema.
- →From an external Python RAG service called over HTTP: ingest the same corpus through Arcana.ingest and move retrieval into your OTP supervision tree, removing the second language and the network hop.
- ↗To a Python framework (LangChain, Haystack): export documents and chunks from Postgres and re-index them in the framework's store when the team leaves the BEAM.
- ↗To a hosted retrieval service: push the same corpus to the vendor's ingestion API and point your app at its search endpoint instead of the local pgvector tables.
Integrations
Resources & guides
Tutorials & Learning
YouTube returned 6 videos for “Arcana”, and we withheld 6: 6 could not be judged, because “Arcana” is a single word that other videos use for other things. We are showing none, because we could not prove any of them are about Arcana.
Official links
What are the best Arcana alternatives?
The closest alternatives to Arcana we list are pgvector, Meilisearch and Vespa. Ranked by direct product-type match, not category overlap.
| Alternative | Pricing | Starts at | Last verified |
|---|---|---|---|
| pgvector | Free | Free | — |
| Meilisearch | Freemium | Pricing not published | — |
| Vespa | Freemium | Pricing not published | — |
Alternatives to Arcana
View all
pgvector
pgvector adds vector similarity search inside Postgres, so embeddings live next to your relational data.
Meilisearch
Meilisearch delivers typo-tolerant, sub-50ms search-as-you-type plus hybrid vector and multimodal retrieval through one REST API.
Tools that pair well with Arcana
Common stack mates teams adopt alongside Arcana, with the specific reason each pairing earns its keep.
pgvector
pgvector adds vector similarity search inside Postgres, so embeddings live next to your relational data.
Meilisearch
Meilisearch delivers typo-tolerant, sub-50ms search-as-you-type plus hybrid vector and multimodal retrieval through one REST API.
Vespa
Vespa is an open-source AI search platform combining vector, text and tensor retrieval with machine-learned ranking in one engine.
Arcana vs its closest rival
Arcana's closest head-to-head is with Spider Cloud — Choose Spider Cloud if you need fast, low-cost web crawling and scraping with AI-powered extraction for any stack. Choose Arcana if you're building RAG inside an Elixir/Phoenix project and want a pure Elixir solution with no external service dependency. Verdict generated
Head-to-head comparisons
Arcana vs Spider Cloud
Choose Spider Cloud if you need fast, low-cost web crawling and scraping with AI-powered extraction for any stack. Choose Arcana if you're building RAG inside an Elixir/Phoenix project and want a pure Elixir solution with no external service dependency.
GeneratedArcana vs Voyage Ai
For Elixir developers wanting an all-in-one, embeddable RAG solution without leaving the BEAM, Arcana is a no-brainer: free, open-source, and deeply integrated with Phoenix and Ecto. But for enterprises needing top-tier retrieval accuracy on specialized domains like finance or legal, Voyage AI's domain-tuned models and low-dimensional embeddings justify its contact-based pricing—especially with the new Voyage 4 series on the horizon. Choose based on your tech stack and need for domain specificity.
GeneratedArcana vs Temporal Ai
Temporal AI is the clear winner for teams needing reliable, fault-tolerant orchestration of AI agents and microservices across multiple languages, with strong support for human-in-the-loop and long-running workflows. Arcana is a niche but excellent choice for Elixir developers seeking to add RAG capabilities to their Phoenix apps without external dependencies. Choose Temporal for multi-language, mission-critical workflows; choose Arcana only if you're deep in the BEAM ecosystem.
GeneratedFrequently asked questions
Used Arcana? Help shape our editorial sentiment research.