v4.5.3
The editor where agents live.
v4.0 is the agent-native author release. The agent goes from "external tool you run a CLI for" to a first-class editor surface — works continuously, across multiple vaults, with full visibility, replay, and rollback. Everything below is in core. No plugin tax. ~15 MB still.
The five pillars
Each is in core. Toggleable via a setting. Built to play nicely with your existing notes — no schema migration, no new file format, no required cloud account.
1 · Inline Agent Panel
Right-side first-class panel (peer to Outline / Backlinks /
Tags / History). Streamed chat-with-vault routed through
the in-process MCP surface and your existing 14-provider
AI stack. [[wikilink]] citations resolve to
real notes; tool-call cards expand inline showing every
read/write the agent makes. Reply too long? Insert
and Copy buttons drop it into the active
editor — replaces selection if any, else inserts at the
cursor. Run history persists as plain markdown under
.solomd/agent-runs/ — grep-able,
git-trackable.
2 · Agent Recipes / Scheduled Runs
Declarative agent jobs as YAML in your vault:
<workspace>/.solomd/agents/*.yml.
Triggers: schedule (cron), on-save,
on-commit, on-tag-add, manual.
Safety rails are non-negotiable: every
run gets its own AutoGit branch agent/<recipe>/<run-id>,
writes are reviewable via accept / reject UI before
merging, and a per-run write-cap (default 5,
hard ceiling 50) prevents runaway loops. Rejected runs
have the branch hard-deleted — zero residue. Ships with
an 11-recipe cookbook (weekly review,
daily summary, TODO extraction, translation, citation
cleanup, CJK proofread agent, …) — install one, edit it,
run it.
3 · Agent Trace View
Every run (Panel chat OR Recipe) emits a
trace.jsonl per step:
prompt / model_call /
tool_call / tool_result /
git_commit / done. Step cards
expand to show args, results, token counts, cost
estimate, AutoGit branch refs.
Replay-from-step button rewinds the run
to step N and re-executes from there with edited inputs —
the path to self-correcting recipes. New MCP tool
read_agent_trace(run_id) exposes traces to
other agents.
4 · Workspace Federation
solomd-mcp --workspace path1 --workspace path2 --workspace path3
— one MCP session, many vaults. Tool signatures gain an
optional workspace parameter; default = first
passed (fully back-compat with single-workspace clients).
AutoGit branches stay isolated per workspace. Settings
→ Integrations adds an MCP profiles UI: name a
bundle of vaults, copy the Claude Desktop config snippet
with one click.
5 · Ollama first-class
We do NOT bundle a local LLM runtime —
Ollama already does that well. v4.0 polishes the
integration: auto-detect at localhost:11434
(green status indicator in Settings → AI), one-click
install hint when not found, three model presets
(rewrite → qwen2.5:7b · quick →
qwen2.5:1.5b · cjk → qwen2.5:14b),
recommended pull from inline UI when no models installed.
Recipes can specify provider: local (alias
of ollama) for cheap autonomous loops with zero cloud
cost. Your notes never leave the machine — true
today via BYOK Ollama; v4.0 makes it discoverable.
🛡️ Safety rails (the bit nobody else does)
- AutoGit branch sandbox. Every recipe run
starts on its own branch. Your
mainstays untouched until you click Accept. A rejected run leaves nothing in your history — branch deleted, file tree reverted. - Per-run write cap. A model that asks to write 50 files when the cap is 5 gets refusals at the 6th tool call. No half-applied state, no surprise mass-edits.
- Refuses to start on a dirty workspace. If you have uncommitted edits when a recipe fires, it errors out cleanly instead of sweeping your work-in-progress into an agent commit.
- Path-traversal hardening. Both the
in-process tool surface and every Tauri command that joins
user-supplied
run_id/patharguments now reject..segments and absolute paths upfront. The agent literally can't write outside its workspace. - Replayable trace. Every step on disk in line-delimited JSON. Suspicious behavior in the panel? Walk the trace, find the bad step, replay from N–1 with edited inputs.
Quality bar
- First-run wizard. New install meets the Agent Panel within 60 seconds — guides you through BYOK key entry OR Ollama setup. Re-launchable from Help.
- Recipe cookbook (11 starters). Settings → Recipes → Browse cookbook. Weekly review, daily summary, TODO extraction, translation pass, citation cleanup, CJK proofread agent, link-rot detector, frontmatter normalizer, outline-to-blog, refactor pass, weekly tag triage. Install one, edit the prompt, run it.
- Public REST API (localhost only, token auth). Same surface as MCP for clients that don't speak MCP yet — Alfred / Raycast / n8n / your own scripts.
- BYOK cost meter. Per-provider running tokens-spent counter, opt-in. Settings → Integrations.
- Localization. ja / ko translations scaffolded for README + iPad metadata. en/zh remain the full-coverage UI languages.
UX polish you'll notice on day one
- IME composition guard everywhere. Pinyin users typing Chinese, Japanese kana mode, Korean Hangul — pressing Enter to commit a candidate no longer fires "send" in chat / "rename" in file tree / "open" in command palette and search. Fixed across 7 input surfaces.
- Insert / Copy on assistant replies. Two buttons under each completed reply. Insert respects the current editor selection (replaces it) or falls back to cursor insertion. Copy is plain clipboard.
- Window UX on macOS. Launch activation timing is right — SoloMD becomes frontmost on cold start instead of opening behind Finder. Restored window position stays inside the current monitor work area; bad saved state from a now-disconnected secondary monitor recenters cleanly instead of opening off-screen.
- Stuck "streaming…" with no error. Fixed. A misconfigured provider (e.g. Ollama with a model that isn't pulled) used to hang the panel forever. Now you see a red error banner with the actual response from the backend, and the Send button comes back.
What v4.0 explicitly skips (and why)
Saying "no" is part of the product. These were considered and consciously not shipped:
- Bundled local LLM runtime. Ollama already covers this; re-implementing violates "write less code we maintain forever." BYOK provider stays the path.
- Online recipe marketplace. Server ops + moderation = off-mission. Cookbook ships in-tree.
- Multi-user / team agents. Violates "one window, one writer." This product is for the single author; collaboration belongs to a different product.
- Copilot-style ghost-text autocomplete. Different brand. Ghost-text dilutes the writer's voice; our agents work at vault granularity instead, with batched reviewable writes.
Upgrade notes
- No file-format changes. Your existing
.mdfiles, AutoGit history, sync setup, BYOK keys all carry over untouched. - Agent Panel auto-enables on first launch.
v3.6.x users had a hidden
showAgentPanel: falsefrom the v4-beta scaffolding period; a one-time migration flips it on so you actually see the marquee feature. Toggle off via⌘⇧P → View: Toggle Agent Panelif you don't want it. - Settings layout adds Recipes and Cost meter tabs under Integrations. Existing settings keep their previous tabs.
- The MCP server (
solomd-mcp) gained a--workspaceflag (repeatable). Older MCP clients with a single workspace argument keep working unchanged; the federation behavior is opt-in.
🔮 What's next
The 5 pillars + quality bar are the headline drop. The v4.x and beyond open directions on the roadmap: a sandboxed scripting API, a CLI v2 (rewrite / commit / semantic search / recipe-trigger from terminal), and the long-deferred "should SoloMD become a business?" question (paid sync — original v3.0 question, still unanswered, no engineering until decided).
Thanks to everyone who filed issues, opened PRs, and ran the v4-beta builds through real vaults during the quiet phase. v4.0 shipped because dogfooding caught the bugs that integration tests didn't. Earlier: browse the blog →