LoginSign Up
← Back to all sparks
O

Ollama

AI-ASSISTANTS
Velocity5.0

Run and manage large language models locally on your own machine.

ollama.com ↗

Ollama doubles as an MLX runtime and a local backend for coding agents

local-llmapple-siliconmlxcoding-agentsllama.cpptool-calling
◆Current state
Ollama is a local LLM runtime maturing on two fronts: a native MLX engine for Apple Silicon, which now runs the Command A and North model families, and an emerging role as a launcher and backend for third-party coding agents, auto-installing Claude Code and opencode and detecting Codex model drift. The bulk of recent tags are release candidates carrying llama.cpp syncs, context-handling fixes, and per-model renderer additions.
◆Where it's heading
Cadence is high but mostly incremental: most tags are RCs bundling dependency bumps and single-model parser work. The directional thread is Ollama positioning itself as the local execution layer beneath external coding agents, alongside deepening MLX support and handling for prompts beyond 8k tokens.
◆Prediction
Expect continued llama.cpp syncs and more launch-provider integrations, with MLX speculative decoding and context-shift work graduating from RC tags into stable point releases.

◆Recent moves

  1. 3d ago

    Tool-call JSON parsing fix; MLX and llama.cpp bumps

    A release-candidate maintenance tag: a tool-call parser fix that ignores braces inside JSON strings, plus MLX and llama.cpp dependency bumps. It fits the steady cadence of upstream syncs rather than any directional move.

    View source ↗
  2. 7d ago

    Adds Ornith 9B renderer and parser support

    Adds renderer and parser support for the Ornith 9B model, the kind of per-model plumbing Ollama ships continuously to keep pace with new model releases.

    View source ↗
  3. 8d ago

    Auto-install Claude Code/opencode; MLX speculative decoding

    The substantive release of this window: auto-install for Claude Code and opencode, Codex model-drift detection, MLX speculative-decoding tuning, and context-headroom fixes. It advances the agent-launcher direction while hardening the MLX engine.

    View source ↗
  4. 15d ago

    Command A and North models run on Apple Silicon via MLX

    Brings the Command A and North model families to Apple Silicon via the MLX engine and bumps llama.cpp to build 9672, incremental progress on the MLX-as-first-class-engine thread.

    View source ↗
  5. 15d ago

    CI: pin Darwin release Xcode version

    A CI-only change pinning the Darwin release Xcode version, with no user-visible effect.

    View source ↗
  6. 16d ago

    Updates bundled llama.cpp engine to build b9672

    A single llama.cpp engine bump to build b9672 ahead of the 0.30.10 release, routine upstream maintenance.

    View source ↗