What you are actually deploying
One desktop application per machine. There is no backend to stand up, no database to host, no container to schedule, no meeting-platform integration to authorise, and no recording bot that joins calls. JesRecap captures the microphone and the system audio through operating-system APIs, runs a speech model locally, and writes the results to the user's own application-support directory.
That means the deployment conversation is a software-distribution conversation, not an architecture project. The parts worth checking are the installers, the network egress, the model files, permissions, and where data ends up.
Status, plainly: JesRecap is on a preview release channel. It is not a shipped 1.0, and there has been no external security audit. If you need audited software with a support SLA, this is not that yet — pilot it, and tell us what a wider rollout would require.
Installers and platform support
| Platform | Requirement | Package |
|---|---|---|
| macOS | macOS 14.2 or newer, Apple Silicon or Intel | .dmg containing a standard app bundle |
| Windows | Windows 10 or 11 | .msi installer |
| Linux | Not available today | — |
Both are ordinary packages, so they go through the tooling you already run —
Jamf, Intune, Munki, SCCM, Chocolatey internal repo, or a file share and an
instruction email. The macOS bundle can be deployed like any other app bundle;
the .msi supports the usual silent-install patterns. Grab the current
builds from the downloads page.
Network surface
Worth reading closely, because it is short. Meeting audio, transcripts, and metadata are never transmitted. The application talks to exactly these places:
| Destination | When | Purpose |
|---|---|---|
models.jesrecap.com |
Once per model, or never if pre-seeded | Plain file download of the speech models, plus the roughly 1.3 GB AI-notes model if that optional feature is enabled. Public mirrors are used as a fallback. |
| Licence API | Activation, and periodic validation | Licence key plus device fingerprint, to enforce the two-device limit |
| Update check | Occasionally | Reads the current published version so the app can tell the user an update exists |
| Stripe checkout | At purchase only, in a browser | Payment. Not a runtime dependency of the app. |
No telemetry, no analytics, no crash reporting to us, no accounts. If you allowlist egress, licence validation is the one call that has to keep working; everything else can be handled once and then blocked. See the security overview for the same list from a security review angle.
Pre-seeding the speech models
The transcription model is a single large file, downloaded once and reused afterwards. Speaker labelling adds a separate one-time 54 MB download. On a rollout of any size you probably do not want fifty machines fetching the same file over a branch link, and on restricted machines you may not want them fetching anything at all.
Both cases are handled the same way: place the model files in the models directory as part of your deployment step, and the application will use what it finds instead of downloading. The directory is shared with JesType, our dictation app, deliberately — if a machine already runs JesType, existing models are reused and nothing needs downloading.
| Platform | Models directory |
|---|---|
| macOS | ~/Library/Application Support/jestype/models/ |
| Windows | %APPDATA%\jestype\models\ |
Practical pattern: download the model once on a staging machine, verify it, host it on an internal share or your management tool's distribution point, and copy it into place at install time. Model files contain no customer data, so they are safe to cache internally.
Storage locations
Everything a user records lives under one per-user directory, which makes backup, retention, and offboarding tractable.
| What | Where |
|---|---|
| Meeting files |
macOS ~/Library/Application Support/jesrecap/meetings/Windows %APPDATA%\jesrecap\meetings\
|
| Library index | meetings.sqlite3 in the jesrecap folder |
| Per-meeting audio | microphone.wav, system-audio.wav, mixed meeting.wav |
| Speech models | The shared jestype/models folder above |
Two things to plan for. First, WAV audio is not small, and both source tracks are kept so meetings can be re-transcribed later — factor that into disk sizing and tell users they can delete audio once a transcript is final. Second, that folder will be picked up by whatever backup you run, so decide whether it should be (encrypted destination, treated as sensitive) or should not be (excluded, and recordings are then not recoverable).
Permissions the user will see
- macOS: microphone access and the system Audio Capture permission, both prompted on first use. macOS shows a recording indicator while capture is active. Pre-approving these through your MDM's privacy preferences policy saves support tickets.
- Windows: microphone access; system audio is captured through WASAPI loopback and needs no special grant.
- Neither platform requires a meeting-platform account, calendar access, browser extension, or admin rights at runtime.
Licensing at volume
Standard licences are one-time — €34.95 for 1 device up to €99.95 for 8 — and cover devices belonging to the same user. There is no subscription and no per-seat renewal. Users can recover a lost key at /recover and manage their activated devices at /manage.
For a team, a department, or a fleet, we would rather talk than publish a table. Tell us the number of machines, the platform mix, and how you want the keys delivered, and we will work out the arrangement — see the enterprise page.
Suggested rollout
- Pilot with two or three people who record meetings often, on the platforms you actually support. Recording quality depends on real hardware and real conference apps.
- Confirm capture works with your stack — Zoom, Teams, Meet, Webex, Slack huddles, or in-person through the microphone. Since JesRecap records the audio the machine already plays, app compatibility is rarely the issue; headset and audio-device configuration sometimes is.
- Fix the models question: pre-seed from an internal share, or allow the one-time download.
- Write the policy before the wide rollout. Who may record, how participants are told, how long recordings are kept, and where exports may go. The GDPR guide and HIPAA guide have checklists you can lift.
- Decide the AI question explicitly. JesRecap has no cloud AI, but its meeting actions have two modes. By default they copy a prompt containing the transcript for the user to paste elsewhere — so name which assistant is approved, or say none is. The optional AI notes on this device setting instead generates the note locally, which keeps transcripts on the endpoint; if you prefer that, say so in the policy and allow the one-time model download.
- Deploy, then keep an eye on disk usage for the first month. It is the only ongoing operational surprise.
Updates
Builds are published on the preview channel and listed on the releases page. The application can check the current published version so users know an update exists; distribution stays under your control, since you are handing out packages through your own tooling. If you need to pin a version across a fleet, distribute that package and skip the newer one.
Questions we expect from a review board
Short answers, so you can paste them into a form: no server component; no telemetry; no accounts; meeting content never transmitted; egress limited to a one-time model download, licence validation, and a version check; data stored per-user on the endpoint; encryption at rest provided by FileVault or BitLocker; no external security audit yet; preview release channel; independent developer, the same team as JesType. Anything that needs more than a sentence — write to support@jesrecap.com and you will get a straight answer rather than a trust-centre link.