Obsidian Copilot vs Smart Connections: features, plans and costs

Obsidian Copilot and Smart Connections are not substitutes — as of Copilot 4.0.9 and Smart Connections 4.7.2 they solve different halves of the job, and the decision that actually costs money is which of four paid tiers you buy. Copilot is now an agent runner: its shipped manifest describes it as "Run AI agents such as Claude Code, Codex, and OpenCode inside your vault". Smart Connections is a link-discovery and semantic-lookup plugin with a bundled local embedding model and, in version 4, no chat surface at all. On that split they are complements, not substitutes.

The part that trips people up is that Smart Chat is three different things with almost the same name, and only one of them ever calls a model API. Everything below was checked on September 19, 2026 against the vendors' own documentation, their pricing pages, the shipped manifest.json of each plugin and the GitHub API. Nothing here was runtime-tested inside Obsidian.

Three plugins, two companies, one confusing name

Versions and dates for the three community plugins come from their shipped manifests — Copilot, Smart Connections, Smart Chat — and the release dates from the GitHub releases API; none of the three repositories is archived. The Pro version numbers come from the vendor's own release pages (Smart Chat Pro v2.2, Smart Chat API Pro v1.4), which carry no dates — those plugins ship through an in-app store rather than GitHub, so treat the Pro line as "current at time of writing" and check in the app.

Smart Connections' README states the split plainly: Smart Chat "has been moved to its own dedicated plugin", and "all API-based model integrations (cloud and local) have moved into Smart Chat Pro". That is the sentence most older tutorials predate. The paid surface also answers to four names — Chat Pro in the bundle list, Smart Chat Pro in the README, Smart Chat API Extension on the website, Smart Chat API Pro on the release pages — so search the in-app store for Chat Pro and expect the other three labels.

Two name traps worth naming before you install. There is an unrelated eugeneyan/obsidian-copilot retrieval prototype, last pushed June 30, 2024, that still ranks for this query and whose setup involves a separate server — it is not this plugin. And "Copilot Plus" here is a Brevilabs tier, unrelated to Microsoft 365 Copilot or GitHub Copilot, which is a different Obsidian plugin again.

Who should pick which

Execution model is the sharpest difference and it is not a feature-list item. Smart Connections reads your vault and suggests; it never writes. Copilot's Agent Chat executes, and its permission control is coarse — the models and parameters documentation gives the opencode backend "Default or Auto permissions" and states that "Plan is not available". The same page says "Copilot V4 does not expose temperature, top-p, or similar tuning in these model lists". If you need a read-only agent that proposes a diff before touching anything, neither plugin is that.

Platform limits cut both products in the same place. Copilot's vault search documentation states that "Agent Chat does not run on mobile"; the Smart Chat FAQ states that "the embedded provider surface, new-chat capture, automatic URL saving, and state actions require the desktop main Markdown pane". Both plugins declare isDesktopOnly: false in their manifests, so they install on a phone — they just do less there.

Migration cost is the line item people forget. Coming from Copilot V3, the in-plugin index is gone: the vault search documentation says "Semantic search from Copilot V3 has moved to Miyo" and that Copilot "no longer exposes controls for its old in-plugin index". Miyo is a separate "Desktop app for Mac, Windows, and Linux" that Copilot connects to — its pricing page says "Have a Copilot lifetime license? Your miyo access is included" — and that page lists Free at $0 (the local path "works without an account or card"), Relay at $4.99 per month and Lifetime at $149.99, where Relay is what adds reaching your local files from remote clients such as ChatGPT and Claude. Coming from an older Smart Connections, chat is a separate install and API models are now paid.

Feature comparison

One security detail worth copying into your own notes: Copilot "stores provider API keys and the Copilot license key in this device's Obsidian Keychain, not in the vault's data.json", per its provider documentation. Syncing the vault therefore does not carry the key to a second machine, which is good hygiene and also the reason a new laptop needs the key entered again.

Obsidian Copilot and Smart Chat pricing: four bills, not one

Both products separate the plugin from the plan from the model bill, and both add a fourth bucket that is easy to miss. Copilot's documentation is explicit: "Free does not mean every model is free. Anthropic, OpenAI, Google, OpenRouter, or another provider may charge your account."

Prices read from Copilot's pricing page and the Smart Plugins pro page on September 19, 2026. Two things this page will not do. It will not translate Copilot's quota into a token number, because no absolute figure is published — the page says only "subject to 5h and weekly quota". And it will not print a Founding Supporter price, because the amount is not in the page text. Confirm both in checkout.

So the honest budget is four lines each. For Copilot: plugin ($0) plus a Copilot plan if you want hosted models, plus Miyo if you want semantic search (free locally), plus your model API bill. For Smart: Core plugins ($0) plus All-access if you want API chat, plus the model API bill — and note that Chat Pro cannot be bought on its own, so the $30 bundle is the entry price for that one feature.

Smart Chat models, and what a custom endpoint can and cannot reach

"Best model for Obsidian Smart Chat" has two different answers, and the split matters more than any model name.

Free Smart Chat does not pick models. You insert a provider codeblock and chat in that vendor's embedded web interface; its README lists smart-chatgpt, smart-claude, smart-gemini, smart-grok, smart-perplexity, smart-deepseek, smart-aistudio, smart-openwebui and smart-kimi, with the generic smart-chat block marked "All providers and API chat (Pro)". Because it makes no API call — "No API key required" is the README's own framing — no gateway is reachable from it, Kunavo included. The only indirect route is smart-openwebui: that block embeds a self-hosted Open WebUI, and Open WebUI is where an OpenAI-compatible base URL would be set. See the Open WebUI integration for that side of it.

Smart Chat Pro documents three providers. The current getting-started flow says to "Select an enabled provider, such as PRO: Open Router (cloud), PRO: OpenAI (cloud), or PRO: Ollama (local, requires Ollama app)", and the Smart Chat FAQ documents the credential fields as an API key for cloud or an Ollama host for local. Neither of those two pages documents a base-URL field. There is countervailing evidence, and it is fair to show both sides: a Custom API (Local or Remote, OpenAI format) adapter exists in the vendor's chat-model source lineage, with separate protocol, hostname, port and path fields rather than one base URL — but on September 12, 2026 a commit titled "Refactor: Move cloud provider adapters to obsidian-smart-env-pro repository" replaced that file with a one-line pointer to external/obsidian-smart-env-pro, and that repository is not public. Kunavo has not bought All-access and has not tested the shipped UI, so this guide asserts neither that it works nor that it does not. Do not plan a Smart Chat migration around it.

Copilot is the opposite case. Its provider documentation describes BYOK → Add a provider → "Add a custom provider", where you "enter its base URL and model ID" and where "The API key is optional; add one when the endpoint requires authentication". Discovery is standard: "Model discovery uses the endpoint's /v1/models response, but you can always enter an exact model ID". That path is not plan-gated — the Free tier is "the complete do-it-yourself path" requiring "no Copilot account, license key, or credit card".

Three boundaries apply to any gateway on that path, and they are the useful part of this comparison.

• Agent Chat coverage is conditional. Copilot's provider documentation says "New chat models are enabled for Quick Chat and opencode by default" and that Copilot "passes the saved key and any custom endpoint override to opencode when it starts" — but also that "A model that opencode cannot route is left out of the opencode list, but may still work in Quick Chat". Worst case a custom endpoint degrades to Quick Chat. That failure mode is the same one OpenCode provider not found covers directly, and the OpenCode integration is the configuration to copy.
• Streaming may be the price of admission. Copilot documents that if Test succeeds but Quick Chat cannot send, you turn on "Enable CORS", after which "Responses then appear after completion instead of streaming token by token". Kunavo's /v1 routes currently emit no Access-Control-* headers and register no OPTIONS handler — verified in our own repository on September 19, 2026. Whether that actually forces the toggle inside Obsidian is inferred, not tested: Obsidian plugins can bypass browser CORS through its requestUrl API, so the real behaviour depends on Copilot's internals.
• Retrieval is out of scope for any gateway. Copilot's provider documentation describes no embedding-endpoint field, and Kunavo serves no embedding model — the semantic-search half of this workflow runs locally through Miyo or Smart Connections' bundled model, not through us.

One more boundary that surprises people: Claude Code and Codex are "Agent Chat backends, not BYOK providers". They authenticate through their own CLIs, so a subscription is not an API key and BYOK settings do not reach them.

A worked estimate for the BYOK path

These are illustrative token arithmetic, not measured task costs and not a bill ceiling. Assume one day of vault work through Copilot's Quick Chat that sends 150,000 uncached input tokens — long notes pulled into context add up fast — and receives 10,000 output tokens. Rates are live Kunavo catalog prices per million tokens.

The reading that matters: under these assumptions Claude Sonnet 4.6 costs $0.240 for that day against $0.080 on Claude Haiku 4.5, so model choice moves this budget far more than any plugin tier does. Comparing that column against $7.99 or $30 per month is not a like-for-like comparison either — a subscription buys hosted capacity and features, while tokens buy exactly the work you did. Cheapest listed rate and cheapest way to finish the task are also different questions; a model that needs three attempts can cost more than a dearer one that needs one. AI cost optimization covers how to measure that on your own vault.

Kunavo's catalog amount is a billing floor rather than a cap: when the upstream reports its charge, the bill is the greater of catalog cost and upstream cost times the applicable markup. Cache charges and external tools sit outside this example. The minimum top-up is $10 in prepaid credit, which is a funding minimum, not a task fee or a subscription — see billing details.

Setting up the Copilot side, honestly

Kunavo has not runtime-tested either plugin, and there is no published Kunavo setup page for Copilot. What is true is narrower and checkable: Copilot documents a custom OpenAI-compatible provider taking a base URL, an optional key and a model id, and Kunavo exposes an OpenAI-compatible /v1 surface including /v1/models, which is what Copilot's discovery reads. OpenAI-compatible API has the base URL and key format, and the OpenCode integration is worth reading alongside it because Copilot's Agent Chat runs opencode underneath. Keep a working route available while you try it, run one bounded task, then read the charge your account actually recorded — and if Quick Chat stalls after a successful Test, the CORS toggle above is the first thing to check. Create a Kunavo account when you are ready to fund a key.

Choosing between gateways rather than plugins? OpenRouter alternatives is the relevant comparison, since OpenRouter is one of the two cloud providers Smart Chat Pro documents, and Copilot's plans documentation names it among the providers that "may charge your account". If the retrieval half is what you actually care about, RAG implementation explains why the embedding step belongs on your own machine.

FAQ

Checked September 19, 2026: both vendors' pricing pages, Copilot's provider, model, vault-search and plan documentation, the Smart Chat getting-started and FAQ pages, the Smart Connections README, all three shipped plugin manifests, the GitHub releases API, and Kunavo's own /v1 routes for CORS headers. Not checked: the shipped Smart Chat Pro settings UI, which is behind a paid subscription, and the runtime behaviour of either plugin against Kunavo. Kunavo token rates come from the live catalog, and every dollar example here is illustrative token arithmetic rather than a measured task cost.