Open-source Perplexica alternatives: the answer engine or the system that runs the work

Perplexica, now shipping as Vane, is a genuinely good open-source answer engine, and it stops at the answer. Kortix is the open-source AI Operating System for the step after it: agents that do the work an answer points at, reviewed before anything lands. This maps the open-source Perplexica alternatives, where each one runs, and the decision that settles the choice, which is whether you need an answer or the work that follows it.

What Perplexica is, and where the rename to Vane lands

Perplexica is an open-source answer engine: it searches the live web, reranks the results, and writes an answer with cited sources, and it runs entirely on your own hardware. The project now ships as Vane, at github.com/ItzCrazyKns/Vane, and its Docker image carries the Vane name while most people still search for it as Perplexica. The project's own release notes still refer to Perplexica by its old name in places, which is how a rename reads in practice (Vane releases).

It reads local models through Ollama or connects to cloud providers such as OpenAI, Anthropic, and Groq, so you choose what answers. SearXNG handles web search, so a query can stay on your machine. The project is MIT-licensed, which lets you fork, modify, and redistribute it. Vane is built around one person and one machine, and its repository still lists authentication among the features it plans to add (checked October 2026).

Perplexity is the closed, hosted engine that inspired it, and you subscribe to it rather than run it. It is the reference point a self-hosted answer engine gets measured against, not one of the open options.

The answer engine, and the work it leaves to you

An answer engine returns an answer. It searches, reads, and cites, and then it stops. That is the whole product, and for one person it is often exactly enough.

A team's work rarely ends at the answer. Someone still sends the reply, opens the fix, reconciles the ledger against the bank, or moves the ticket. An engine that only writes answers leaves every one of those steps to a person, and it holds no shared memory, no roles, and no review step, so two people working the same problem each start from zero.

None of that is a flaw in an answer engine. Questions and answers are the job it was built for, and it does that job well. The gap opens when the job changes from answering to doing.

Kortix: the open-source system you own

Everything a team's agents need is a file in Kortix. Agents, skills, company memory, connector config, and triggers sit in one git repo the company owns, and the repo is versioned and diffable, so you can grep the whole company or roll any part of it back (Kortix on GitHub).

Every session runs on its own isolated cloud computer on its own branch. The agent can install, run, and break anything, and only what it commits survives. Session work reaches the main branch through a change request a person reviews, and merge is default-deny for agents, so nothing lands unreviewed.

Kortix connects to 3,000+ apps plus any MCP, OpenAPI, GraphQL, or raw HTTP API, and connector credentials are brokered server-side so the raw key never enters the machine (kortix.com). You pick the model per agent, per session, or per message with your own API keys, or sign in with the ChatGPT plan you already pay for. Run it on Kortix Cloud, in your VPC, or on your own on-prem network. Self-host is free, and the managed plan is a flat per-seat monthly fee (Kortix pricing).

The unit that matters here is the layer: an answer becomes a task, the task runs on a machine of its own, and the result arrives as a diff a person approves.

The other open options

Vane (Perplexica). Reach for it when one person needs private, cited answers over the live web. It is free to self-host, MIT-licensed, and bundles its own search engine in a single Docker image, so your only cost is the model you connect (Vane).

Open WebUI. A self-hosted AI interface built to run entirely offline, connecting to Ollama or any OpenAI-compatible API, with roles, groups, LDAP, SSO, and retrieval over your own documents (Open WebUI). It is strong at the chat layer and reaches the web through many providers, SearXNG among them. Its memory is personal by default, so what it learns does not become shared company knowledge, and it ships under the Open WebUI License, a permissive licence whose branding clause keeps the name visible unless a deployment covers 50 or fewer people in 30 days or holds written permission (Open WebUI License).

Khoj. A self-hostable second brain that answers from the live web and from your own documents, including PDF, markdown, Notion, and Word files, and runs on local or online models with custom agents and scheduled research (Khoj). It scales from a personal install up to a cloud or on-prem deployment, and it is open source and self-hostable by design.

The local stack, SearXNG plus Ollama. SearXNG is a metasearch engine that aggregates results from other search services and tracks or profiles nobody (SearXNG). On its own it returns links, not a written answer. Ollama runs open models on your own hardware. Pair the two behind a front end and you have an answer engine assembled entirely from parts you run, at the cost of standing up and maintaining each part yourself.

Side by side

When to pick which

Pick Kortix when you want the company's agents, skills, memory, and connector rules in one repo you own, each session on its own isolated machine, and every change landing as a diff a person approves. You choose the model per agent with your own keys, and the free tier costs nothing to start.

Pick Vane when one person wants private, cited answers on their own hardware with no server to run. Pick Open WebUI when the work lives in conversations and you want a self-hosted chat layer with real identity and permissions. Pick Khoj when the job is a searchable second brain over your own documents. Pick the SearXNG plus Ollama stack when you want to assemble the engine yourself and keep every part open. Pick Perplexity when you want managed, hosted answers and no infrastructure to run.

An answer engine and a system that runs the work are two layers more than two rivals. Start where the work is, and move up a layer when the answer stops being the deliverable.

If a piece of this needs a deeper read, open-source Perplexity Computer alternatives covers the closed digital-worker category, the best self-hosted open-source AI workspace compares six self-hosted team platforms, and Kortix vs Open WebUI takes one pairing all the way down.

Free to start, free to self-host. Try Kortix and run your first agent from a repository your company controls.