v0.3.2

Persome Runtime v0.3.2

Persome v0.3.2 adds the metadata and executable entry point required for the
Official MCP Registry. The Runtime remains local-first: Registry clients launch
the same persome mcp stdio server from the public personal-model PyPI
package, and personal data stays on the owner's Mac.

Install or upgrade

uv tool install personal-model
# Existing installs:
uv tool upgrade personal-model

persome onboard

The personal-model executable is an alias for persome, allowing MCP
Registry installers to launch personal-model mcp directly.

v0.3.1

Persome Runtime v0.3.1

Persome v0.3.1 adds the first official PyPI distribution and keeps the existing
source-update path compatible with v0.3.0 installations.

Install from PyPI

uv tool install personal-model
persome onboard

The PyPI distribution is named personal-model; the installed command remains
persome. persome onboard performs provider setup, explains and requests the
effective macOS permissions, verifies bundled OCR when enabled, starts the
Runtime, and requires a mode-appropriate fresh capture/readiness receipt.

Upgrade a PyPI installation with its package manager and re-run Runtime proof:

uv tool upgrade personal-model
persome onboard

Source installation remains supported:

git clone https://github.com/Intuition-Lab/personal-model.git
cd personal-model
bash install.sh

Publishing and compatibility

• PyPI publication uses GitHub Actions Trusted Publishing with the protected
  pypi environment; no long-lived PyPI token is stored in GitHub.
• The PyPI wheel and source distribution are built from the same tagged source,
  smoke-tested outside the checkout, and published only after the GitHub Release
  job succeeds.
• The root source project retains its persome-core compatibility identifier so
  the already released v0.3.0 updater continues to accept new source checkouts.
  Runtime code recognizes both distribution metadata names.

All v0.3.0 onboarding, owner-local API protection, transactional updates,
model-recovery hardening, local viewer, MCP, and privacy guarantees remain in
this release.

v0.3.0

Persome Runtime v0.3.0

Persome v0.3.0 narrows the public Runtime around local capture, durable personal
memory, the inspectable Point/Line/Face/Volume/Root model, and MCP. It replaces
the early alpha's implicit setup assumptions with explicit onboarding, stronger
owner-local security, and transactional updates.

Highlights

• repeatable persome onboard permission, OCR, daemon, and fresh-capture proof;
• provider-aware LLM setup for Anthropic, OpenAI-compatible services, Azure, and
  local endpoints, with completion and tool-calling checks before saving;
• owner-local bearer protection for REST, viewer, and HTTP MCP routes, plus
  origin, request-size, and local-runtime hardening;
• transactional persome update with lifecycle restoration and rollback after
  failures or interrupts;
• safe repair of derived SQLite/FTS indexes without deleting personal-model
  state, plus temporal-history preservation during rebuilds;
• redesigned local model viewer, dense synthetic showcase, and an explicit
  model-share card that never publishes without the owner's action;
• cleaner stdio MCP lifecycle and isolated test roots;
• removal of the legacy terminal Chat surface so the standalone Runtime stays
  focused on capture, memory, modeling, and agent access through MCP.

Install

git clone https://github.com/Intuition-Lab/personal-model.git
cd personal-model
bash install.sh

persome status
persome model open

Existing installations can update transactionally from any directory:

persome update

The GitHub Release includes a source distribution, wheel, and SHA256SUMS.
install.sh remains the supported first-run path because it compiles the macOS
AX helpers, establishes private local state, runs provider setup, and proves the
permission and capture path. The same wheel and source distribution are the
artifacts published to PyPI.

Verification

The release workflow verifies the locked dependency graph, hash-constrained
build backend, offline test suite with mocked LLMs, ruff, shell syntax,
secret/PII/language/documentation scans, synthetic MCP transport, clean wheel
installation outside the checkout, bundled Swift/Three.js/PP-OCRv6 resources,
checksums, and GitHub artifact provenance.

Compatibility notes

• Live capture requires macOS 13 or newer.
• Apple Silicon supports bundled local OCR; Intel Macs keep AX capture, Runtime,
  and MCP but do not receive the Paddle OCR runtime.
• The legacy persome chat command and browser Chat routes were removed. Connect
  Claude Code, Codex, Claude Desktop, Cursor, opencode, or another client through
  persome mcp instead.
• Except canonical GET /health, local HTTP surfaces now require the
  owner-generated API token or the one-use viewer capability. Use
  persome model open instead of opening /model directly.
• There is still no hosted account, sync service, telemetry, meeting-audio
  capture, or remote deployment contract.

v0.2.2

Persome Runtime v0.2.2

This is the first public alpha of the standalone Persome Runtime: a local-first
macOS daemon that collects focused cross-app context, builds an inspectable
Point/Line/Face/Volume/Root personal model, and exposes it to Chat and MCP
clients.

Highlights

• macOS Accessibility capture with optional, bundled on-device PP-OCRv6;
• five-minute active-session modeling without stopping the daemon;
• durable Markdown plus SQLite/FTS5, receipts, bitemporal history, correction,
  redacted export, and complete deletion;
• streamable HTTP and stdio MCP, terminal Chat, and a local 3D model viewer;
• synthetic sample model and real MCP transport verification with no key or
  personal data;
• source distribution, wheel, and SHA256SUMS release artifacts.

Install

git clone https://github.com/Persome-ai/persome-core.git
cd persome-core
bash install.sh
persome start
open http://127.0.0.1:8742/model

The GitHub Release wheel is a verified install artifact, but install.sh is
the supported first-run path because it also compiles the macOS AX helpers,
creates the private environment file, and configures local clients.

Verification

The release workflow runs the offline test suite on mocked LLMs, ruff, shell
syntax, secret/PII/language/documentation scans, synthetic MCP transport smoke,
wheel installation outside the checkout, and bundled-resource checks. Verify
downloaded artifacts with either platform's built-in checksum tool:

# macOS
shasum -a 256 --check SHA256SUMS

# Linux with GNU coreutils
sha256sum --check SHA256SUMS

Known limitations

• Live capture supports macOS 13 and newer. Linux is for offline development;
  Windows is unsupported.
• Intel Macs support AX capture, Runtime, Chat, and MCP, but not bundled OCR
  because the required Paddle runtime has no compatible Intel wheel.
• Accessibility permission is required for useful live capture. Screen
  Recording is optional and used only for local OCR fallback or encrypted
  screenshot retention.
• Semantic modeling requires a user-supplied Anthropic API key or compatible
  gateway. Capture and BM25 retrieval continue without one.
• The loopback API has no second bearer-token layer. Connect only trusted local
  clients and do not expose the port through a tunnel.
• This alpha has no hosted account, sync service, telemetry, browser Chat UI,
  remote deployment contract, or PyPI publication.
• Synthetic Runtime gates are not a real-person memory-quality or next-action
  benchmark. Research evaluation belongs in a separately licensed benchmark
  repository.