What is PARTITA STUDIO

PARTITA STUDIO combines an LLM-based chat agent with a full-featured MIDI editor. You state your musical intention — "a heroic theme in D-minor for full orchestra" — and the system automatically:

• Plans the structure and arrangement
• Generates parts for each instrument, taking into account articulations, dynamics, and range
• Processes the result through a MIDI pipeline (quantization, harmonization, humanize)
• Sends the final MIDI to your DAW via a virtual MIDI port

Key Features

Workflow

A typical workflow in PARTITA STUDIO looks like this:

1. Create a project — select a template or start from scratch, set tempo, key, and time signature
2. Add instruments — choose from the profile library or create your own
3. Define structure — create blocks (intro, verse, chorus, etc.) and assign chords
4. Describe your idea — the agent plans the arrangement and generates parts
5. Refine the result — edit notes in the Piano Roll, rewrite individual bars from a description, play a part in yourself on a MIDI keyboard, or ask the agent to revise
6. Balance it — volume, pan and solo/mute in the mixer
7. Export — save the MIDI file or send notes directly to your DAW

System Requirements

Installation

Windows

1. Download the installer Partita-Setup-X.Y.Z.exe from the homepage
2. Run the installer — it may require admin privileges to install to C:\Program Files\PARTITA STUDIO
3. You can select a different installation folder if desired
4. PARTITA STUDIO will launch automatically once finished

macOS

1. Download the universal Partita-Setup-X.Y.Z.dmg image from the homepage
2. Open the image and drag PARTITA STUDIO into the Applications folder
3. Launch the app from Applications; the build supports both Apple silicon and Intel Macs
4. The current build is not yet signed or notarized by Apple. If macOS blocks the first launch, try opening it once, then go to System Settings → Privacy & Security, scroll to Security, and click Open Anyway. Only override this warning for the image downloaded from partita.pro.

Updates

PARTITA STUDIO checks for updates automatically on launch and every 4 hours. When a new version is available:

1. A notification showing download progress will appear
2. Once downloaded, a "Restart and update" button will be shown
3. The update will be installed on the next launch or when clicking the button

You can also check for updates manually: Settings → Development → Check for updates.

A note on macOS: the current macOS builds are not signed with an Apple certificate, so the system does not allow in-place updates. The app will announce a new version and offer to download the installer — grab the DMG and reinstall over the old copy; your projects and settings are kept. This is expected behaviour, not an error.

DAW Setup

To send real-time MIDI to your DAW, use the LoopMIDI virtual MIDI port on Windows or the built-in IAC Driver on macOS. Read more in the DAW Connection section.

Step 1. Connect a Model

PARTITA STUDIO runs on your models and your keys. Open Settings (the gear icon in the top right) → the Agent section:

1. Pick a provider: OpenAI, OpenAI Codex, Anthropic, Gemini, OpenRouter, Polza.ai, Ollama or LM Studio
2. Paste the API key. You can also go keyless: the OpenAI Codex provider signs in with OAuth through your ChatGPT account, and Anthropic accepts a token from claude setup-token
3. Click the model field — the list loads by itself — and choose a model
4. Go to the Generation section and set up the second slot the same way

You can stay off the cloud entirely: Ollama and LM Studio run models locally and need no key — just the address of your local server.

Step 2. Create a Project

Press Ctrl+N or the New Project button in the toolbar. Choose a template:

• Orchestral — a full orchestra (strings, brass, percussion)
• Chamber — a smaller ensemble
• Empty project — add instruments manually

Set the tempo (BPM), key, and time signature in the timeline top bar.

Step 3. Describe Your Idea

Open the chat (Ctrl+L) and tell the agent what you want to hear:

The agent will then:

1. Analyze the current project state
2. Search the knowledge base for orchestration best practices
3. Plan the arrangement (which instruments play what and when)
4. Generate the parts for each instrument
5. Verify the result and make adjustments if necessary

Step 4. Listen and Refine

Press Space to play back. If something is off, say it in plain words — there is no need to redo the project:

• "Make the violin melody more expressive" or "Remove the trumpets from the first four bars"
• "The cello is off the harmony in bar 12 — fix just that note"
• "Slow down towards the ending: from bar 29 ease down to 80 BPM"
• Right-click a clip → Send to chat to make the agent work on that fragment specifically
• Or open the clip in the Piano Roll and edit the notes by hand

The agent can see what is already written: it reads the whole score of a block, checks the vertical bar by bar, and edits in place instead of rewriting everything.

Step 5. Export

When you're happy with the result:

• MIDI file — use the "Export MIDI" button in the toolbar
• To DAW in real-time — enable WebMIDI in settings and press Play
• Save Project — save as a .partita file to continue working later (Ctrl+S)

Where to Go Next

The five minutes are over — this is where the interesting part starts:

• Coherence & Score — how to make the parts sound like one ensemble and bring themes back
• Precise Editing — change one note, one CC or the tempo of three bars without regenerating
• Tempo & Meter — accelerations, ritardandos and meter changes inside a piece
• Instruments & Profiles — connecting your own libraries and articulations
• DAW Connection — routing the parts into Kontakt and other samplers
• Troubleshooting — when something does not sound, or sounds wrong

General Structure

The PARTITA STUDIO interface consists of several zones:

Toolbar

The top panel contains:

• Project Management — name, new, open, save, export MIDI (for the whole project or specific blocks)
• Global Parameters — BPM, key, time signature (can also be changed via InfoBar on the timeline)
• Generation (Magic Buttons) — quick generation in the selected block: for the entire arrangement, melody, accompaniment, bass, or a single clip
• Undo/Redo — undo and redo actions
• Chat — toggle the AI agent panel
• Settings — open application settings

Timeline

The central area is the visual representation of your project:

• InfoBar (top) — global parameters, block addition, revision history
• Blocks — horizontal sections (intro, verse, chorus, etc.). Each block has its own chords and length
• Lanes — instrument tracks inside each block
• Clips — cells at the intersection of a lane and a block, containing generated notes
• Mini previews — each clip can show the overall note shape directly on the timeline
• Curve Overlay — overlay of macro curves (intensity, momentum, space) over the blocks
• Ruler — displays bar numbers indicating the current position

A context menu (right-click) on blocks and clips provides access to operations: rename, delete, duplicate, send to chat, etc.

Chat Panel

The right panel housing the AI agent (toggled with Ctrl+L):

• Multi-sessions — several independent chats
• Attach clips — send a specific part to the agent for refinement
• Tool calls — shows which tools the agent is using
• Thinking — reveals how the agent "thinks" (for reasoning models)
• Diff cards — summary of the changes made by the agent

Clip Editor

A floating panel above the arrangement (double-click on a clip or press 4):

• Piano Roll — full C0–B8 range, draw, select, delete notes
• Velocity Lane — edit the dynamics of each note
• CC Lane — MIDI controller curves (CC1 Modulation, CC11 Expression, etc.)
• Articulation Lane — articulation switch points
• Sustain Pedal — toggle the sustain pedal

Floating panels

The editor, the chat and the mixer are capsules — they sit above the arrangement instead of taking space from it for good:

• Height (editor and mixer) is dragged from the top edge, chat width from the left edge
• The expand button stretches a panel across the workspace, Esc brings it back
• When the chat is open the other capsules narrow automatically — panels never overlap each other

Transport Bar

The bottom playback control strip:

• Play / Pause / Stop and record
• Current position (time and measure)
• Loop — loop the selected section
• Metronome — click to toggle, right-click for count-in and level settings
• Magnet — snap seeking to the timeline grid
• MIX — the mixer with volume, pan, mute and solo
• Generation and error indicators

Creating a Project

A new project is created via Ctrl+N or the button in the toolbar. The following options are available:

• From a template — a pre-filled instrument set and structure for common scenarios
• Empty project — a blank canvas where you add instruments and blocks manually
• From custom template — your own saved templates

File Format

Projects are saved in the .partita format — a JSON file containing the full project state: instruments, blocks, chords, generated parts, and curves. .partita files are associated with the application and open via double-click.

Saving

• Ctrl+S — save the current project
• If the project hasn't been saved yet — a file picker dialog opens
• An unsaved changes indicator appears in the title bar

Opening

• Ctrl+O — open a .partita file
• Double-click a .partita file in Explorer
• Recent projects list in the new project dialog

Revisions

PARTITA STUDIO automatically saves project revisions. You can revert to any previous version using the revision history button on the timeline InfoBar. Revisions are stored in the application's data directory and linked to the project ID.

Project Templates

You can save your current project as a template for future use. The template saves instrument settings and block structure (without generated notes).

What is a Block

A block is a section of your composition with a specific length (in bars), type, chord progression, and a set of clips for each instrument.

Block Types

The block type affects its visual display and acts as a hint for the AI:

• intro — introduction
• verse — song verse
• chorus — song chorus
• bridge — connecting section
• outro — ending
• development — section evolving materials
• transition — transition between parts
• climax — peak of the piece
• custom — user-defined type

Block Operations

• Adding — "+" button on the InfoBar or via context menu
• Moving — drag-and-drop on the timeline
• Duplicating — copies the block along with all its clips
• Changing length — duration in bars is set in the block's properties
• Inheriting — a block can inherit a theme from another block (for thematic coherence)

Phrase Patterns

Every block can have a phrase pattern — an outline of how musical phrases are distributed across bars. This pattern guides how the AI builds melodic lines:

• AABB — two groups of 2-bar phrases
• ABAB — alternating phrases
• ABAC — phrases with contrast at the end
• Custom patterns

Swing and Groove

A block can carry a swing setting — the delay applied to offbeats that produces the characteristic lilt of jazz, shuffle, and many pop grooves. You choose the depth (from a straight 50% up to an extreme 75%) and the grid it applies to: eighths or sixteenths.

Swing is stored with the block and reapplied on every generation, so regenerating a part never loses the feel. You can also just ask — "give it a light sixteenth-note swing" — and the agent sets it for you.

The model still writes notes on a straight grid; the feel is applied afterwards during processing. That keeps swing predictable and identical across every part in the block.

Adding Instruments

Click the add instrument button on the timeline. A profile browser partitioned by instrument family will open:

• Strings — violins, violas, cellos, double basses (CSS, Nucleus, Spitfire)
• Brass — horns, trumpets, trombones, tuba (Nucleus, CSB)
• Woodwinds — flute, oboe, clarinet, bassoon (CSW)
• Percussion — orchestral and epic percussion (Strezov, GM)
• Keys — piano, harp

Instrument Profile

Each profile includes:

Custom Profiles

You can create and edit instrument profiles:

• Via AI — ask the agent in the chat: "Create a profile for Spitfire Chamber Strings"
• Manually — the profile editor (right-click → "Edit Profile") lets you set all parameters yourself
• Custom profiles are saved in the application's data directory

Instrument settings

Right-click a track → Settings. The dialog is split into three parts, each holding only what belongs to it:

• Instrument — the profile, its range and articulations
• Sound — the source (built-in or external instrument) and everything that depends on it
• Level and colour — track volume and its colour on the timeline

Sound source

The source is chosen per track and always wins over the global playback mode:

• External instrument — the part goes out to a MIDI port (LoopMIDI, IAC Driver) and plays in Kontakt, PLAY, Sine or another sampler. Port and MIDI channel are set here
• Built-in — the SoundFont bank or the profile's own synthesizer. Timbre and channel are set here

So some parts can go to a sampler while the rest play back internally. When a track's choice disagrees with the global mode, the dialog says so — that way "the bass went silent" reads as "this track is deliberately different from the project". Tracks whose source was never set follow the global mode.

The GM program only appears where it changes something: external instruments never receive Program Change (it would reset Kontakt patches), and a profile with its own synthesizer sets the timbre itself.

MIDI Routing

• Channel — MIDI channel (1–16)
• Port — MIDI output for tracks with an external source
• Auto-assign — PARTITA STUDIO can distribute channels automatically

Record mode

Each track has its own live-recording mode — replace or overdub. Right-click the ● button to switch it; see MIDI Keyboard Recording for details.

Chord Editor

Chords are defined per-block. Open the chord editor via a block's context menu on the timeline. Each chord is anchored to a specific bar.

Chord Formats

Standard musical notations are supported:

• Basic triads: C, Dm, E7, Fmaj7
• With modifiers: Gsus4, A♭dim, B♭aug
• Inversions (slash chords): C/E, Am/G

Progression Presets

PARTITA STUDIO includes a library of ready-to-use progressions categorized by mood:

• Classical (I-V-vi-IV, ii-V-I, etc.)
• Cinematic
• Dramatic
• Modal

The agent can also suggest a progression if you describe the feel you're going for.

Impact on Generation

During generation, the AI sees the chord map for the block, noting the exact pitches for every chord — e.g., Bar 1: Dm (D, F, A). This assists the model in crafting melodic lines that fit the harmony seamlessly and placing chord tones accurately on downbeats.

The Three Curves

Editing

The curves are displayed as an overlay on top of the timeline (toggled by the curve visibility button in the toolbar). Each curve consists of breakpoints with values between 0 and 1. The AI agent can automatically draw macro curves when planning the arrangement.

Base Tempo

A project has one overall tempo (BPM), set in the top panel, and it applies wherever nothing else is specified. If that is all you need, you can stop reading here: an empty tempo map means the old behaviour.

The Tempo Map

On top of the base tempo sits a map: a list of spans in the form "from this bar through that bar — this tempo". Bars are counted from the start of the piece, not from the start of a block, so a request like "80 BPM from bar 29 through 32" means exactly that.

Playback and Export Agree

The player and the MIDI export are built from the same tempo map. Smooth ramps are written out as frequent steps, so the ritardando you hear in your DAW is the one you heard in the app.

Meter

The meter is set per block, and where needed per bar — a single 2/4 bar in the middle of 4/4, for instance, to keep a phrase from sprawling. Compound meters (6/8, 9/8, 12/8) are respected both when strong beats are analysed and when the vertical is checked.

How the Agent Works

The agent operates via an interaction loop with the language model:

1. You send a message (a prompt, a request, a question)
2. The agent builds a payload with a system prompt detailing its role and the tools available
3. The model replies with text and/or tool calls
4. The agent executes the tools locally, gets the results, and hands them back to the model
5. The loop keeps running until the model outputs its final answer without calling any new tools

The agent streams text back to the UI in real time, so you can read along as it generates.

Operating Modes

Multi-sessions

You can manage multiple independent chats within a single project. Each session keeps its own history and context. Switch between them using the tabs in the chat header.

Attaching Clips

For pinpoint adjustments, send a particular clip into the chat:

1. Right-click on a clip in the timeline → "Send to chat"
2. The clip appears as an attachment card above your text input
3. Tell the agent what to fix — taking the exact notes in that clip into account

Task Queue

You can queue up multiple requests while the agent is busy. The queued requests will be executed one after another.

Reasoning Effort

For models with configurable reasoning, a level selector is available: off, low, medium, high, and for some models also xhigh. That covers the Claude 4 and Claude 5 families (Fable, Opus, Sonnet, Haiku 4.5), OpenAI's GPT-5.x line and o-series (including via OpenAI Codex), and Gemini 2.5 and 3.x. Claude 5 models additionally support an adaptive mode, where the model decides how much to think on each step. A higher reasoning tier enhances planning quality at the cost of longer response times.

Context and Compaction

The agent monitors the size of your conversation. When the context length edges near the model's limit, the app invokes automatic context compaction — summarizing older messages while preserving essential semantic cues.

Generation Pipeline

Each instrument part goes through a multi-stage pipeline:

1. Context — gathering information: instrument profile, block chords, neighboring parts, macro curves, block boundaries
2. Prompting — forming a query to the LLM with rules specific to the instrument type (melodic or percussive)
3. Generation — the LLM returns JSON containing notes, articulations, and curves
4. Parsing — the JSON is parsed into a GeneratedPart structure
5. Processing — the MIDI pipeline applies a series of transformations

MIDI Pipeline

After being parsed, each part passes through a sequence of transformations:

Selective Context

When generating an instrument, you can dictate which other parts the model receives as context:

• All — the model sees all existing parts in the block (default)
• Selective — only specified instruments (for instance, a pad only sees the melody)
• No Context — the part is generated completely standalone
• Cross-block — feeding parts from other blocks to ensure thematic unity

Notes appear while generating

While the model writes a part, notes show up in the clip — the first ones usually within a couple of seconds, then the part fills in as you watch. This is a preview: it writes nothing into the clip, it exists so you can tell early whether the model went the right way and stop a bad take without waiting for the end.

The finished result is always applied whole, from the complete answer: quantization, range, overlap resolution and dynamics are computed across the entire part at once. That is why an interrupted generation is an error rather than "half a written clip".

The clip the composer is working on carries an indicator: a frame in the instrument colour and a soft animation over exactly the bars being rewritten.

Takes

You can generate multiple variations (takes) for a single clip. Quickly switch between them to pick the best result. Takes are accessible via the clip's context menu.

How it works

1. Hold Alt and drag across a track — bars inside a single block get selected
2. A capsule with a prompt field appears next to the selection. Describe what those bars should do
3. Press Generate — only the selected bars are rewritten

Material outside the range survives by construction: the model's result is spliced into the original at the seam, and notes crossing the boundary are cut exactly on it. Even if the model drifts beyond the brief, it cannot leak past the selection.

Other ways in

• Shift+Alt+W — open the capsule for the selected clip as a whole
• Clip context menu → Improve or redo — the same thing by mouse
• The capsule's "To chat" button — when the edit is easier to discuss with the agent than to state in one line

When the range covers a whole clip, the part is written as a normal generation — a "don't touch bars outside the range" instruction is meaningless for the entire clip.

Empty bars

You can select empty space too: if no clip exists there, one is created exactly on the selection and written from scratch. "Put a four-bar fill here" is a normal request, not an error.

Context

The toggle in the capsule decides whether the composer sees the neighbouring parts of the block. With context the fragment sits better in the arrangement; without it, it follows your description more literally.

What is an Arrangement Plan

An arrangement plan (ArrangementPlan) is a structural object formed by a single LLM call for each block. It consists of:

• Vision — a high-level summary of how the block should sound
• Directives — bespoke instructions for each instrument, defining bounds like role, articulations, entry measure, dynamics, and what context they need to hear

Instrument Directive

Every planned instrument receives:

• Role — its function in the ensemble (melody, harmony, bass, rhythm, countermelody)
• Description — a descriptive briefing of the part
• Entry bar — the measure when it starts playing
• Articulations — which playing techniques to employ
• Dynamics — the specified dynamic range
• Context instruments — which other parts to "listen to" during generation

What's Inside

71 documents filled with highly practical knowledge:

Beyond orchestral topics, the base covers modern arranging: pop song form, EDM builds and drops, hip-hop beat construction, kick and bass interplay, drum kit programming with ghost notes, synth roles, and modern keyboard voicings.

How the Agent Uses Knowledge

During the planning phase, the agent queries the knowledge base via the query_music_knowledge tool. The search supports filtering by category, sample library, instrument family, and genre. The retrieved guidelines guide the agent in selecting appropriate articulations, voicings, and stylistic maneuvers.

When the Agent Searches

The knowledge base answers questions about music, but it does not know what shipped last week or what a particular library's page says. Search closes exactly that gap: current information, fact checking, and reading documentation at a link you provide.

The agent decides on its own and only reaches out when it makes sense. You can also ask directly — "search the web", "check the developer's site" — or simply paste a link.

What Actually Leaves Your Machine

Search runs through the Parallel.ai service. Only the search query the agent composed, or the address of the public page to read, is sent. Your project, notes, messages, and model API keys are never transmitted.

Local network addresses are rejected before anything is sent — localhost, private IP addresses, and links carrying a username and password cannot be fetched.

Turning It Off

Settings → Agent → Web search. With the switch off, the search tools are not offered to the model at all, so no query can leave by accident.

How Memory Works

Memory is a collection of records that the agent persists between sessions. Typical examples include:

• "The user prefers CSS legato for violin melodies"
• "Use Strezov X3M percussion for trailer music projects"
• "Avoid overly dense arrangements in quiet sections"

Managing Memory

The agent stores, searches and deletes records itself — you can simply say "remember that I work in 6/8 and dislike heavy brass" or "forget my percussion preferences". Memory lives in the application data directory and is available in every project.

The Score of a Block

Before changing anything, the agent reads a map of the block: one row per part, one column per bar. Each cell shows the register and the number of attacks, a sustain mark across the bar line, or silence. Separate rows carry the harmony (with borrowed chords marked) and the energy.

That view immediately shows what a single clip hides: a part that is silent for half the block, three instruments entering at exactly the same moment, an energy dip where the climax was supposed to be.

Checking the Vertical

When a specific spot needs attention — "what is going on in bar 19" — the agent takes a vertical slice: the chord of the bar, the notes of every instrument, how well they fit the harmony, and non-chord tones highlighted on strong beats. That is cheaper and sharper than reading every part in full.

Ensemble Rules

The classic part-writing problems are checked separately:

• Parallel fifths and octaves between the outer voices — two voices moving the same way at the same interval, sounding glued together
• Voice crossing — the sustained kind, not a passing one: a part that stays above or below its neighbour against the usual order
• Register conflicts — parts sitting in the same band, so nothing is distinguishable

The Compositional Intent

The agent keeps a separate record of what is actually being composed: the form and the energy arc, the central motif and its transformations, the register lanes of the instruments, the agreements about transitions between blocks.

That record survives edits, rollbacks and context compaction. The practical consequence is simple: the theme stated in the introduction comes back in the finale instead of dissolving halfway through.

What the Composer Sees

When a part is generated, the model is shown only what it needs rather than the whole project: the melody for a counter-line, nothing at all for an ostinato, a part from another block as a reference for a call-and-response. Less irrelevant context means fewer chances the model starts "reacting" to unrelated voices instead of doing the job.

What a Skill Does

• Adds instructions to the agent's prompt — orchestration rules for trailer music, for example
• Brings its own knowledge documents, which join the shared search
• May expose settings you can change in the interface

Managing Skills

Settings → Skills. Bundled skills ship with the app; user skills are installed from a folder or from a link to a manifest.json. Each skill is switched on and off separately, and its card shows the author, version, tags and contents.

Your Own Skill

The easiest route is to ask the agent: "build a skill for jazz arranging". It will lay out the structure, and you fill in the instructions and knowledge. A finished skill is an ordinary folder — you can hand it to a colleague.

Overview

PARTITA STUDIO's Piano Roll is a high-performance canvas-based editor spanning the complete C0 to B8 range (108 keys). Open it by double-clicking on any clip or pressing the 4 key.

Tools

Navigation

• Mouse wheel — horizontal scroll
• Shift + wheel — vertical scroll
• Alt + wheel — vertical zoom
• Ctrl + wheel — horizontal zoom
• When you open a clip, the editor automatically scrolls to frame the existing notes

Lanes

Positioned below the main note grid are supplementary lanes:

• Velocity — dynamic bars for each note (0–127)
• CC — MIDI continuous controller curves (CC1 Modulation, CC11 Expression, etc.)
• Articulations — points marking articulation switches, carrying text labels
• Sustain Pedal — toggle events for the sustain pedal

Snap-to-Grid

Snap to grid is enabled by default. The finest step is a 32nd note. Snapping applies to drawing, moving, and stretching notes.

Why It Matters

The usual trouble with generative tools: to fix a detail you have to run generation again — and the detail changes together with everything you liked. Here an edit is addressed: the agent changes exactly what you asked for, and the rest stays byte for byte the same.

Addressing by Bar and Beat

A note can be named musically: "bar three, beat two, the D". That kind of address survives notes being inserted or deleted around it, and tolerates humanisation — a note nudged slightly off the grid is still found.

Grids Down to 1/64 and Triplets

The step of an edit is set by the grid — from a quarter down to 1/64, triplets included (1/8T, for example). The grid also decides how dense CC points should be: a gentle curve does not need an event on every tick.

What Comes Back

After every edit the agent reports what actually changed: which bars were touched and the values before and after. There is no need to re-read the whole clip to confirm the result — and it is what protects you from silent edits that missed their target.

Recording a pass

1. Connect a MIDI keyboard. The device is picked in Settings → Audio & MIDI; by default every input is listened to
2. Press ● on the track you want — it arms. From that moment you hear your playing through that instrument, before recording anything
3. Put the playhead where you want to start and press record in the transport (or Alt+R)
4. A metronome count-in plays (1 or 2 bars, configurable), then recording starts
5. Stopping the transport commits the pass into the clip. If no clip exists under the playhead, one is created

Takes

Every pass is stored as its own take of the clip — the previous version is never lost and you can switch between them. Record as many passes as you like and keep the best one.

Overdub and replace

Right-click the ● button to choose the record mode for that track:

• Replace (default) — the previous content moves into a take and the pass is recorded from scratch. During recording the old material is muted so you only hear yourself
• Overdub — the pass is added to what is already there, without a new take. The clip keeps sounding: playing on top only works if you hear the foundation. Handy when a part is built in layers — chords first, then bass, then ornaments

Controllers are not blended: if you moved modulation or expression during the pass, older events for that same controller inside the recorded span are replaced. Two automation lines on one CC would not sound like "both" — they would sound like jitter.

Quantization

Settings → Audio & MIDI lets you pick the record quantization grid: off, 1/4, 1/8, 1/16 or 1/32. Quantization snaps note starts to the grid but keeps the release point, so durations stay alive. Anything already recorded can be tightened separately with Q.

Metronome

The metronome button in the transport: click to toggle (C), right-click for its settings panel — count-in length (0, 1 or 2 bars), downbeat level and off-beat level.

The click follows the project meter: 5/4 and 7/8 click on the notated beat, compound meters (6/8, 9/8, 12/8) click on dotted quarters, so 6/8 gives two clicks per bar. The tempo map is respected, so the metronome slows down together with the music.

Built-in SoundFont

Out of the box, PARTITA STUDIO uses a built-in General MIDI synthesizer built on a SoundFont engine. This gives you instant playback without requiring additional setups. While the sound quality is rudimentary, it caters well to evaluating melodic and rhythmic concepts.

WebMIDI

For professional-grade sonics, route PARTITA STUDIO to external virtual instruments (like Kontakt, PLAY, Sine, etc.) using WebMIDI. Check out the DAW Connection section for more details.

Per-track sound source

The playback engine is not a single global switch. Every track picks its own source in the instrument settings: built-in (the SoundFont bank or the profile's own synthesizer) or external instrument (a MIDI port). An explicit per-track choice always wins over the global mode, so some parts can go to Kontakt while the rest play back internally. Tracks whose source was never set follow the global mode.

Controls

• Space — Play / Pause
• Click on the ruler or an empty part of the timeline — set playback head position
• Loop — continually cycle through the selected region
• Magnet in the transport — snaps seeking to the timeline grid. On by default: a click on the ruler lands exactly on a beat instead of "roughly there"
• Metronome — click to toggle, right-click for its settings (see MIDI Keyboard Recording)

Playback Position

The playhead is indicated on both the timeline and the Piano Roll. If stopped, the position stays exactly where it paused; pressing Play resumes from there. To return to the very beginning, hit Stop.

Latency

The right end of the info bar shows a latency indicator: MIDI input and audio output separately, colour-coded green, yellow, red. The metronome panel has a slider to compensate for it. The click is scheduled on the same audio clock as the music, so the metronome adds no latency of its own — what remains is the shared output latency, identical for the click, the parts and live playing.

The mixer

The MIX button in the transport (or F3) opens the mixer — a floating panel over the arrangement. Every track gets a cell tinted with its instrument colour: a vertical volume fader, a pan control filled from the centre, and M / S buttons. Muted and non-soloed tracks dim in the mixer and on the timeline alike, so their state reads from either screen.

The header has "Unmute all" and "Clear all solo". The panel height is draggable from its top edge and is remembered; the expand button stretches it across the workspace and Esc collapses it again.

Volume

Volume is set in decibels and applied live — you can ride a fader while the project plays. Click the value to type an exact number.

Pan

Pan spreads parts across the stereo field. The built-in synthesizer is panned by a real stereo node; external instruments and the SoundFont bank receive CC10, so the pan travels to your sampler along with the notes and behaves the way your library interprets it.

Format

PARTITA STUDIO exports Standard MIDI Files (SMF) Type 1 at a 480 PPQ (ticks per quarter note) resolution. Each instrument gets its own dedicated MIDI track, incorporating:

• The MIDI channel defined in the instrument's routing
• Keyswitch notes or CC events used for articulations
• Continuous controller curves (expression, dynamics)
• Tempo markers

How to Export

1. Click the "Export MIDI" button on the toolbar
2. Choose a destination directory and filename
3. Drop the resulting .mid file into your DAW

How to Import

Drag one or more .mid files straight into the chat panel. They are parsed, stored with the project and attached to your next message — the agent sees their tracks, notes and tempo.

What to Do With It

• "Take the melody from this file and arrange it for strings"
• "This is a sketch from my DAW — build the harmony around it and add a counter-line"
• "Use the drum rhythm from the second track as the basis for an ostinato"

What You'll Need

• LoopMIDI on Windows
• The built-in IAC Driver in Audio MIDI Setup on macOS
• A DAW: Cubase, Reaper, FL Studio, Ableton Live, Studio One, etc.
• Virtual instruments (Kontakt, PLAY, Sine, etc.)

Configuration

1. On Windows, install LoopMIDI; on macOS, open Audio MIDI Setup and enable the IAC Driver. Create a port named "PARTITA STUDIO"
2. In PARTITA STUDIO: go to Settings → Audio & MIDI and select that virtual port as the output
3. In your DAW: create MIDI tracks and select the same virtual port as their MIDI input
4. Assign MIDI channels in your DAW to match PARTITA STUDIO's routing

Put simply: create a virtual MIDI port with the tools provided for your system, then select it in both the app and your DAW.

It doesn't have to be all or nothing

You don't have to push the whole project into a sampler: the sound source is chosen per track (Instrument settings → Sound). Strings can play in Kontakt while scratch drums and a pad stay on the built-in bank — and the other way round, a single part can be sent out without moving the rest.

Program Change is never sent to external tracks: libraries like Kontakt don't use it, and an incoming message would reset the loaded patch. Instead CC1 and CC11 are initialised on connect — otherwise an instrument sitting at zero dynamics stays silent even when it receives notes.

Reconnecting

PARTITA STUDIO automatically detects when MIDI ports disconnect and reconnect, for example after restarting a virtual port or waking from sleep. If a port is lost, it attempts automatic reconnection with exponential backoff. The port status indicator (green/red) is visible in the MIDI settings.

Two Model Slots

PARTITA STUDIO uses two independent slots. Every request goes straight from your machine to the provider you chose, with your key — the app does not proxy it through our servers.

How the Roles Are Assigned

Four roles work inside those two slots, and how they are assigned matters for quality:

The form and the plan come from the Agent slot — this is one of the most audible places where cutting costs costs music. The notes come from the Generation slot, and there a fast model usually does the job.

• Economy pairing — a strong model in the Agent slot, a fast one in Generation. A sensible balance for everyday work
• Maximum quality — a strong model in both slots. Noticeably more expensive, but you hear it in phrasing and detail

Supported Providers

The model list is not baked into the app: it is fetched from the provider when you open the model field, so new model versions show up on their own, without updating PARTITA STUDIO.

Which Models to Pick

Guidelines as of August 2026, matched to the two slots:

Reasoning Level

A reasoning-level switch appears next to the model — but only for models that support it: Claude 4/5 with extended thinking, OpenAI's GPT-5.x and o-series, Gemini with thinking, and reasoning models via OpenRouter and Polza.ai. The levels are off, low, medium, high, with xhigh on some models; Claude 5 also offers adaptive, where the model decides how much to think. A higher level helps noticeably with planning and review, and barely matters while notes are being written.

Configuring a Slot

1. Pick a provider from the dropdown menu
2. If you use an external cloud provider, enter the API key
3. Specify a Base URL only when it is actually needed (local models, self-hosted setups, or proxies)
4. Click the model field — a pop-up will fetch the available models
5. Select your preferred model

Audio Output

Select the device that will process playback from the built-in SoundFont synthesizer. It defaults to your system's primary output.

MIDI Port

Choose the target MIDI port for broadcasting WebMIDI data. PARTITA STUDIO enumerates all available MIDI outputs on your system. In practice, that usually means LoopMIDI on Windows or IAC Driver on macOS.

Port status indicator:

• ● Green — port connected and engaged
• ● Red — port disconnected or encountered an error

The "Test" button sends a brief note (C4, 300ms) to verify the connection.

MIDI Input

Settings for recording live performance:

• Device — which MIDI keyboard to take notes from. By default every input is listened to, with an availability indicator next to it. Devices are picked up on the fly, so a keyboard can be plugged in without restarting the app
• Monitoring — hear your playing through the armed track
• Record quantization — off, 1/4, 1/8, 1/16 or 1/32

Recording itself is covered in MIDI Keyboard Recording.

Trial Period

Every new user gets 30 days with full access to every feature. No card is required to sign up. The trial limits neither the number of projects nor exporting — it is the complete application, just for a while.

Signing In with an Email Code

Your email is confirmed with a six-digit code from a message — the code is right in the subject line, you enter it on the same screen where you signed up, and it logs you straight in. The same works for signing in without a password: request a code from the sign-in screen. It is also the only route if you forget your password — there is no separate reset procedure. A sign-in code is valid for 10 minutes, a confirmation code for 30. The message comes from auth@partita.pro — if it is not in your inbox, check the spam folder and add the address to your safe senders.

A Licence, Not a Subscription

You buy the application itself. The models and the keys are yours, so you pay your AI provider directly rather than paying us for access to someone else's tokens.

Payments go through YooKassa. Buying early does not burn the time you already paid for: the new term is added to whatever remains.

How the Instalment Plan Works

• One charge per month. Every paid month counts toward the buyout, and the counter is visible in your dashboard
• After the 36th payment the licence becomes lifetime automatically and the charges stop
• The plan cannot be lost. It cannot be cancelled — only paused, for as long as you need
• While paused both the charges and the access stop, and the days you paid for but did not use are saved and returned when you resume
• Pausing does not reset the buyout progress
• You can detach the saved card in the dashboard: charges stop, the paid period and buyout progress are kept, and the plan continues with a new payment that links a card again

Dashboard

Account management is available in the dashboard on the website and inside the app's settings (Account section): licence state, buyout progress, payment history and totals.

If a Payment Fails

Both the app and the dashboard show a plain-language reason and what to do next: retry later, update the payment method, or resume a paused instalment plan. If a payment is stuck as pending, refresh its status in the dashboard — there is no need to pay again.

Project and editing

Playback and generation

Panels

MIDI editor

Mouse on the timeline

• Alt + drag across a track — select bars and open the regeneration capsule
• Ctrl + drag a clip — copy it to another track or another block
• Clip edges — drag them to change its length

Arranger Templates

Built-in templates provide standard orchestral setups and structures for various genres. A typical template incorporates:

• A pool of instruments along with their profiles
• A block framework mapping out form sections
• Advised chord progressions
• Recommended instrument routing and roles

Alongside the orchestral setups there are modern ones: Pop Verse-Chorus (verse, pre-chorus, and chorus with a rhythm section and keys), EDM Build & Drop (a rise and drop built on synths and a drum machine), and Trap Beat (808 bass, hats, sparse texture).

Custom Templates

You can package your active project as a template via the toolbar menu. Custom templates save the following:

• The instrument lineup and assigned profiles
• Block outlines (types, lengths, chords)
• Global project variables (BPM, key, time signature)

Note that generated MIDI data is not saved inside templates — solely the architectural foundation.

Cinematic Studio Strings (CSS)

String profiles for solos, sections, and ready-made ensembles with detailed CC-mapped articulations:

• Violin, 1st Violins, 2nd Violins, Viola, Violas, Cello, Cellos, Bass, Basses
• Lite Ensemble and Full Ensemble presets for faster setup
• Articulations: legato, sustain, spiccato, staccato, tremolo, pizzicato, col legno, con sordino, marcato, and more
• Articulation mode: CC (CC58 sequences)

Cinematic Studio Brass (CSB)

Brass profiles for solos, sections, and grouped setups:

• Solo Horn, Solo Trumpet, Solo Trombone, Solo Bass Trombone, Solo Tuba
• Horn, 4 Horns, Trumpet, 2 Trumpets, Trombone, 2 Trombones, Tuba, Full Ensemble
• Articulations: sustain, staccato, marcato, sforzando, fortepiano

Cinematic Studio Woodwinds (CSW)

Woodwind profiles for single lines and small sections:

• Solo Piccolo, Solo Flute, Solo Alto Flute, Solo Oboe, Solo Cor Anglais, Solo Clarinet, Solo Bass Clarinet, Solo Bassoon, Solo Contra Bassoon
• Flute, 2 Flutes, Oboe, 2 Oboes, Clarinet, 2 Clarinets, Bassoon, 2 Bassoons
• Intricate articulation arrays and expression mapping

Nucleus (Audio Imperia)

A compact orchestral library supplying larger ready-made sections:

• Large strings and brass: 16 Violins, 10 Violas, 6 Celli, 4 Double Basses, 6 French Horns, 3 Trumpets, 4 Trombones, 2 Tubas
• Grouped Brass, Low Brass, and paired woodwind setups

Strezov (X3M)

Epic and traditional orchestral percussion setups:

• Orchestral Percussion — timpani, snare drums, piatti, tamtam
• Epic Percussion — taiko, grand gran cassa, aggressive blockbuster percussion

General MIDI

The standard palette configured for the built-in SoundFont engine. Ideal for swift compositional prototyping absent of external libraries.

Modern Instruments

Profiles for pop, electronic, and rock music, written for any synth or sampler rather than one specific library:

• Synths — Lead, Pad, and Pluck driven by filter and resonance (CC74/CC71) instead of articulations
• Bass — Sub 808 with a long glide, and Electric Fingered
• Drum machines — 808 and 909 mapped to the General MIDI layout
• Keys — electric piano and organ
• Guitar — Electric Clean

These profiles need no articulation switching: expression comes from velocity and continuous controllers, so the agent writes for them differently than for orchestral libraries.

Reading this section is not required: the agent picks its own tools. It is here if you want to make sense of the tool-call cards shown under the agent's replies.

Capabilities Load on Demand

The agent does not keep the entire toolset in front of it. It opens the group it needs — editing, or generation, say — when it gets to that kind of work. This saves context and leaves more room for the music itself.

Reading

Writing

Knowledge, Memory and Skills

Searching the knowledge base, working with long-term memory, enabling and configuring skills.

Silent WebMIDI Playback

Agent is Unresponsive / API Error

Kontakt Ignore Articulation Switches

Notes Breaching Range Constraints

Application Fails to Update

The Verification or Sign-in Code Never Arrives

Diagnostic Logs

For more granular debugging insights, inspect:

• Generation Log — %APPDATA%/partita/Partita/logs/generation_debug.log
• Agent Log — saved on a per-session basis to your local storage
• Settings → Developer — displays direct paths and buttons to reveal your logs in Explorer

Where to Write

• support@partita.pro — support: account access, licences, payments, application bugs
• sales@partita.pro — commercial enquiries, team and educational licences, partnerships
• auth@partita.pro — the service address codes are sent from. Mail sent to it is not processed
• Telegram channel — news and releases