0.1.8

Improved

• Separate notes cited in an answer from context notes retrieved but not cited. Number cited notes in first-mention order and keep the remaining notes in a collapsed list.
• Use the same citation numbers and source grouping when copying a chat as Markdown.

Fixed

• Recognize square and decorative citation markers, including spacing and invisible characters inside the brackets. Show unknown source markers as unverified.
• Restore normal line wrapping when a provider inserts narrow no-break spaces between words. The saved provider response is preserved.

Earlier release history is available in GitHub Releases.

Full Changelog: 0.1.7...0.1.8

0.1.7

What’s new

• Added a model picker for each chat-provider profile.
• Refresh available model IDs from the provider’s OpenAI-compatible /models endpoint using the API key stored in Obsidian SecretStorage.
• Preserved manual model-ID entry for providers without discovery support and retained configured models that are absent from discovery results.
• Shared provider URL validation between model discovery and chat completion requests.

Full Changelog: 0.1.6...0.1.7

0.1.6

What's Changed

Fixed

• Numeric settings no longer persist bad values — blank or invalid entries in numeric settings (e.g. searchLimitPerVault, maxNotes, maxContextChars) are validated on blur and discarded instead of being saved to your config.
• Sessions are protected while a response is streaming — creating or deleting a chat session mid-response is now blocked with a visible notice, so an in-flight generation can't corrupt session state.
• Streaming edge cases — OpenAI-compatible responses that end without a trailing newline are now flushed correctly, and cancelling a response rejects cleanly as an abort.

Improved

• Chat session management rework with tighter type boundaries across the retrieval and client modules.
• OhsMcpClient now takes an explicit version parameter for safer service negotiation.
• Minor layout polish: composer spacing and message font size.

Documentation

• README gained a retrieval-tuning guide (candidate results / notes read / context capacity presets, e.g. balanced 24/10/32,000 vs. high-recall 32/12/48,000), privacy details, and a chat window screenshot.

Verification

• 71/71 tests passing; bundle verified free of dynamic code execution; release assets attested via GitHub Actions with SHA-256 checksums.

---

Full Changelog: 0.1.5...0.1.6

0.1.5

What’s changed

Retrieval improvements

• Retrieval now searches only the current question. Previous conversation turns remain available to the answer-generation model but no longer influence search results.
• Added optional one-hop traversal of explicit links and backlinks for relationship-oriented questions.
• Related-note traversal remains vault-local, is disabled by default, and does not run for ordinary factual searches.
• Cached OHS MCP capability discovery to reduce repeated setup overhead.

Interface improvements

• Linked notes are identified in the source list.
• Sources remain collapsed after successful retrieval but open automatically when retrieval fails or returns no results.
• Improved spacing below the chat composer.

Upgrade notes

• The former query-expansion setting has been removed.
• Enable One-hop related notes under Hybrid Chat’s retrieval settings if you want graph-assisted relationship searches.
• Existing endpoint, vault, and provider settings remain compatible.

Full Changelog: 0.1.4...0.1.5

0.1.4

What's new

• Added configurable retrieval query expansion: Off, Follow-ups only (default), and Always.
• Added bounded conversational variants for referential follow-ups and optional lexical variants for broader recall.
• Centered long-note context on the matching passage and backfilled unreadable top-ranked notes from lower-ranked candidates.
• Restored the last-used vault scope and selected endpoints when reopening the chat view.

Privacy and compatibility

• Older OHS endpoints that do not advertise queries[] continue to receive only the original query.
• Copy actions write Markdown to the system clipboard only after an explicit user click. Hybrid Chat never reads clipboard contents.

Full Changelog: 0.1.3...0.1.4

0.1.3

Fixed

• Removed the transitive AJV runtime code that introduced dynamic code execution in main.js, replacing the bundled MCP SDK transport with a small typed Streamable HTTP client.
• Restricted release attachments to the three files supported by the plugin installer: main.js, manifest.json, and styles.css. Checksums remain internal to build-provenance attestation.
• Added a release gate that rejects generated bundles containing eval() or new Function().

Full Changelog: 0.1.2...0.1.3

Hybrid Chat 0.1.2

Hybrid Chat 0.1.2 completes the repository and manifest compliance fixes required for community plugin submission.

Fixed

• Added an MIT license with copyright © 2026 Jens Mittelbach.
• Removed the redundant reserved word from the manifest description; it now reads: “Federated, provider-neutral RAG chat over existing hybrid-search MCP services.”
• Declared the MIT SPDX license in package metadata.
• Added regression coverage for both the manifest-description restriction and repository license metadata.

Version 0.1.2 retains the valid hybrid-chat plugin ID introduced in 0.1.1.

Full changelog: 0.1.1...0.1.2

Hybrid Chat 0.1.1

Hybrid Chat 0.1.1 corrects the plugin manifest ID so the release passes Obsidian manifest validation.

Fixed

• Changed the plugin ID from obsidian-hybrid-chat to hybrid-chat, which follows the current manifest rules and is not present in the community registry.
• Renamed the internal command ID to open-chat; the user-visible command remains Hybrid Chat: Open chat.
• Updated package metadata, compatibility metadata, the MCP client version, and the documented manual-installation folder.
• Added regression coverage for the manifest ID character and reserved-word restrictions.

Upgrading from 0.1.0

Version 0.1.0 cannot pass current Obsidian manifest validation because its ID contains obsidian. If you installed it manually, rename the entire plugin folder from .obsidian/plugins/obsidian-hybrid-chat/ to .obsidian/plugins/hybrid-chat/ so its data.json settings move with it, then enable Hybrid Chat again.

Full changelog: 0.1.0...0.1.1

Hybrid Chat 0.1.0

Hybrid Chat 0.1.0 is the first public release of the desktop-only Obsidian plugin for federated, provider-neutral RAG over existing Obsidian Hybrid Search services.

Highlights

• Search the current vault, selected vaults, or all enabled OHS endpoints concurrently.
• Merge per-vault rankings with reciprocal-rank fusion and retrieve only the globally selected notes.
• Stream answers from OpenAI-compatible providers with API keys stored through Obsidian SecretStorage.
• Preserve conversational context for follow-up questions while keeping assistant answers out of retrieval queries.
• Show namespaced citations with current-vault and cross-vault navigation.
• Surface partial endpoint failures, support cancellation and persisted sessions, and export chats as Markdown to the clipboard.
• Apply exact YAML property filters and optional current-vault property context through explicit @property(...) directives.

Privacy and scope

Hybrid Chat creates no index, reads no OHS database directly, and does not modify vault notes. Retrieval uses only the OHS MCP search and read tools. A loopback chat provider can keep generation local; remote providers receive the packed source context described in the README.

Requirements and installation

Requires Obsidian 1.13 or newer on desktop. Download main.js, manifest.json, and styles.css into <vault>/.obsidian/plugins/obsidian-hybrid-chat/, then enable Hybrid Chat in Obsidian.

The accompanying checksums.sha256 file can be used to verify the release assets.

Full changelog: https://github.com/jmiba/obsidian-hybrid-chat/commits/0.1.0