What it does
5 features
Pivra Desk · Personal agent
Product and access
Pivra Desk: separate product profile from the personal-agent research.
Core features
5 features from provider sources · checked on 16 October 2026. Availability and limits are listed below.
Usable AI models
Open choice
Without an account, a small Gemma or Qwen model runs on your computer. Alternatively, you add a key from Anthropic or OpenAI, use OpenRouter, or an OpenAI-compatible endpoint, including Ollama or LM Studio. Included cloud credits are announced but not yet on sale; until then, cloud models need your own key.
Model families named
- Google (Gemini)
- Alibaba (Qwen)
- Anthropic (Claude)
- OpenAI (GPT)
- Other models can be connected freely
- Model choice
- You choose
- Your own API key
- Yes
- Router or custom endpoint
- Yes
- Local models
- Yes
The rating covers the generally available main access. Documented · Sources accessed on
Pivra Help: Choosing a model · Pivra Desk · Pivra Desk: Pricing
Does it fit my task?
2 practical tasks · 3 requirements and limits
A good fit if …
- you want to run an agent on a Mac with Apple silicon or Windows 10/11
- you want to use a local model or your own API key
- risky actions such as sending or deleting should run only after your approval
Less suitable if …
- routines must also run while the computer is off or asleep
- you need the paid plans, which are currently only on a waitlist
Editorial assessment based on provider documentation. Not a product test.
Practical tasks
Editorial examples based on documented features. No product tests performed.
Editorial starter task
- Your input
- Draft a comparison table with source references from three local quotation files.
- Possible result
- A reviewable draft with open questions and the inputs used.
- What to check
- Compare the result with the inputs; approve external actions only after personal review.
Example: Prepare research
- Your input
- A bounded task for “Prepare research” with approved sample data.
- Possible result
- Search the web and files for relevant information.
- What to check
- Compare the result with the inputs; approve external actions only after personal review.
Draft a comparison table with source references from three local quotation files.
Your possible prompt
Organize these shared documents by topic. Create an overview with deadlines and next steps. Prepare draft replies for my approval.
Adjust the task, data and permissions.
Requirements and limits
Can I use it?
With requirements
Europe & regional availability
The local deployment path is specifically documented. Regional cloud availability remains unresolved.
Reviewed scope: Documented self-hosting; cloud and external services assessed separately
Reviewed on:
With requirements
Local deployment with prerequisites
The free local deployment path is documented: Apple Silicon Mac or Windows 10/11, without an account and optionally with a local model. Alternatively, your own Claude or OpenAI API access is required; its regional conditions apply separately.
Region, reason and source
Region covered: Your own environment in Europe
Documented reason: Technical deployment path subject to device and model requirements; no blanket availability of external services.
Availability still unclear
Europe: specific availability unconfirmed
The free desktop version uses local models or your own API key; paid plans are waitlisted. European purchase and cloud availability remain unresolved.
Region, reason and source
Region covered: Germany, EU/EEA, United Kingdom and Switzerland
Documented reason: The reviewed sources identify no regulatory reason for a regional restriction.
Accounts & permissions
Based on provider documentation. Permissions depend on the account, integration and organization policies. Review the permissions actually requested before use.
Configuration covered: Free desktop app on your own computer, local model or your own API key
Sources reviewed:
Show all 3 points
Documented with conditions
Account used
In offline mode with a local model the product page says you need neither an account nor a key. You use cloud models with your own key. Google sign-in asks only for calendar (read-only) and account email, GitHub uses the device flow, and remote MCP servers sign in with OAuth. The product page lists one-click Google and GitHub sign-in as a Plus feature, and Plus is not yet on sale. Keys and tokens are kept in an encrypted local vault.
Evidence · EvidenceOpen in reviewed sources
Admin approval
The reviewed pages describe per-action approvals by you, but no approval or restriction by administrators. According to the pricing page, every member of a Pivra Pro or Business workspace gets Desk Plus; what administrators control there is not stated.
Evidence · EvidenceDocumented
Data access & team sharing
Files stay where they are or in an assigned work folder, and the browser uses its own profile. On macOS commands run in a sandbox by default without access to home folders. The page describes no team sharing; only the channel owner can answer Slack and Teams cards.
Evidence · Evidence
Memory and conversation continuity
1 of 6 criteria documented
Local memory database; automatic saving also depends on the model.
Desk 0.2.26, local Mac/Windows app
Sources accessed:
Documentation by criterion
StorageDocumented
Storage and scope
Chats, memory, rules and routines reside in a local database in the app's data folder.
SelectionPartly documented
Selection and timing
Automatic memory depends on the model: it stays off with Qwen2.5 0.5B. Exact selection rules and write timing are unspecified.
Edit & deletePartly documented
Inspect, edit and delete
Deleting the data folder removes local data. Inspecting, editing or deleting individual memories is not explained in these help pages.
Access boundariesPartly documented
Agent and channel boundaries
Data stays on the individual computer; agents have separate browser profiles. This does not establish how memory is shared between agents.
RetrievalPartly documented
Retrieval
Cloud models receive the context needed for a call. Selection of stored memories for that context is not publicly detailed.
New conversationsPartly documented
Session reset and compaction
Completed runs lose full message history after one week; the answer and trace initially remain. Memory compaction is not documented.
Cost and effort
1 plan · Prices as of 16 October 2026
Prices and included usage
Pricing checked
Free: USD 0; Plus listed at USD 6/month. Sales availability conflicts.
| Plan | Price | Included |
|---|---|---|
| Free | USD 0 | One agent, your own or local model, files, routines and approvals. |
- The pricing page offers Plus at USD 6 monthly; the Desk landing page still puts paid plans on a waitlist. Plus + Credits at USD 15 monthly is announced. Model costs for your own keys are separate.
Setup
Apple Silicon Mac or Windows; local model or own API key.
Operation & oversight
Based on public provider documentation. Assessments are editorial inferences from the stated properties. Actual time requirements and reliability have not been measured.
Deployment covered: Local on macOS or Windows, operating while the app is running
Sources reviewed:
Editorial assessment
Operation depends on your computer staying on, and risky actions come to you for approval. You should watch routines and the activity log yourself, since failure notifications are not described.
Show all 6 points
Documented
Files and working state
Agents, chats, runs, memory, rules and routines are in a local database in the data folder on your computer. Finished runs lose their full message history after a week; runs, approval cards and notifications are deleted after 90 days.
EvidenceOpen in reviewed sources
Sign-ins
The reviewed pages give no validity period for connections. If the macOS keychain is locked, saved logins are unavailable until it is unlocked.
EvidenceDocumented with conditions
Starting further runs
Routines run only while the app is open and the computer is on, not while it is asleep or shut down. A routine missed during sleep runs once after waking. Your own routines are on once saved; template routines stay off until you tick them.
EvidenceDocumented
Human intervention
Reading and searching run automatically. Sending, publishing, commands and changes outside the work folder wait for your approval. Money, deletion and deployments require Touch ID or a typed phrase. If you do not respond, the action does not run.
EvidenceOpen in reviewed sources
Detecting errors
Finished and unfinished runs are listed under While you were away and in Activity. The help page does not describe error details or failure alerts.
EvidenceOpen in reviewed sources
Resuming after failures
The reviewed help page does not describe retrying or resuming after failures. Pivra Cloud for always-on work is planned as an opt-in.
Evidence
Evidence and open questions
3 sources · 2 open points
Open points
The reviewed sources contain no reliable information on these points. A gap says nothing about the product.
Europe: Availability still unclear
Requirements and limits: not yet confirmed
Sources and evidence
Provider sources reviewed; the following examples are editorial suggestions, not hands-on tests.
Editorial team: Michael Meierhoff · sources reviewed on . Provider information is not independent evidence of performance.
Provider
Provider of Pivra Desk; identity and features documented through linked sources.
- Operator
- Pivra
- Organization
- Provider
- Website
- www.pivra.ai
Changes to this profile 1
Added Pivra Desk with provider sources, access status and boundaries.
These notes describe editorial changes. Source and pricing review dates appear in their respective sections.