Skip to content

v0.1.0 Synarmo - Context-aware auto-suggest engine

Latest

Choose a tag to compare

@vrraj vrraj released this 10 Jul 22:26

Synarmo - Context-aware auto-suggest engine - Platform v1

Synarmo Platform v1 establishes the project as a local-first auto-suggest
platform: a reusable Python engine, a local service layer, a browser-based
interaction surface, and a documented GGUF model setup path for private
type-ahead suggestions.

This release turns Synarmo from a package-shaped prototype into a platform that
can be embedded, served locally, tested without a model, and run with real
local llama.cpp inference.

Highlights

  • Local-first suggestion engine with SynarmoEngine.load(...),
    engine.suggest(...), and convenience prediction APIs for embedding
    Synarmo in Python applications.
  • Backend-swappable model layer with a deterministic mock backend for
    tests and wiring checks, plus a llama-cpp backend for local GGUF inference.
  • Short type-ahead suggestions tuned around next-word and next-phrase
    completion rather than broad chat responses.
  • Local REST and WebSocket service mode for applications that need a warm
    local Synarmo process instead of in-process Python calls.
  • Browser /ui experience for interactive local suggestion testing,
    context comparison, and autocomplete evaluation.
  • Model readiness workflow through synarmo model-ensure and
    make model-ensure, including local cache handling and configured model
    download support.
  • Environment-driven runtime configuration for model cache, model filename
    or path, suggestion count, token limits, temperature, top-p, logprob pool,
    GPU offload layers, and llama.cpp verbosity.
  • GPU setup and diagnostics for llama.cpp inference, including
    SYNARMO_N_GPU_LAYERS, CPU-only fallback, supported offload checks, and
    runtime diagnostics surfaced by the CLI and service health endpoint.
  • LAN-capable local service binding with synarmo serve --host 0.0.0.0
    and make serve-lan for trusted-network testing.
  • Developer Make targets for service startup, mock startup, browser UX,
    model checks, health checks, and shutdown.

Platform Surface

Platform v1 includes four supported ways to use Synarmo:

  • Python API for direct embedding through SynarmoEngine.
  • CLI for quick local suggestions, interactive compose flows, model checks,
    and service startup.
  • Local service for REST and WebSocket clients that need a persistent warm
    engine.
  • Browser UI for local evaluation and hands-on type-ahead testing.

Documentation

The README and companion docs were reorganized around the v1 platform path:
real local inference first, browser UX setup next, shared model configuration
in one place, then API, service, architecture, and roadmap details.

This release also clarifies:

  • when a GGUF model is downloaded or only verified;
  • how LOCAL_MODELS_CACHE, SYNARMO_MODEL, and SYNARMO_MODEL_PATH are used;
  • how mock mode differs from real llama.cpp inference;
  • how to configure CPU-only vs. GPU-offloaded inference;
  • how to run the local service on loopback or a trusted LAN;
  • how suggestion controls such as choices, candidate tokens, candidate words,
    temperature, top-p, and logprob pool map to runtime behavior.

Not Included

  • No model files are committed.
  • No generated profiles or personal conversation data are included.
  • UI applications beyond the local browser surface remain outside the reusable
    core package.