Skip to content

Release notes

0.3.0

  • Added OpenAIModel, AnthropicModel, and GoogleGenAIModel on the official provider SDKs.
  • Added normalized provider settings, tool calls, usage, safe error codes, cancellation cleanup, and optional dependency guidance.
  • Persisted complete assistant tool calls before tool results and migrated SQLite to the current schema.
  • Added provider-aware muru init --provider starters and executable provider examples.
  • Added provider extras with verified SDK floors and aligned Python, frontend, workflow, and distribution versions.
  • Rebuilt the documentation around Get started, Build, Providers, Operate, Reference, Labs, and Project tasks.
  • Clarified that the Python package and browser Workspace are the MVP while native work remains in Labs.

Limitations: use one active runtime process per SQLite file. The package does not include PostgreSQL, durable trace export, a tool sandbox, or production identity management. Real provider access and quotas remain deployment-specific.

0.3.0-alpha.1 native preview

  • Added a separate native terminal workspace, hardware discovery, responsive layouts, durable redacted events, and session restore.
  • Added signed model-catalog verification, atomic installation, compatible llama.cpp runtime selection, authenticated loopback supervision, constrained decisions, and bounded residency.
  • Added portable agent-pack contracts, deterministic requirements compilation, simulation-only routing, explanations, and measured fixture gates.

This remains a separate GitHub prerelease. The public model catalog is empty and action effects remain simulated.

0.2.0

  • Added SQLitePersistence for sessions, messages, runs, ordered events, artifacts, approvals, and idempotency.
  • Added WAL, foreign keys, schema versioning, atomic event counters, bounded busy retries, and stable storage errors.
  • Added honest restart recovery with process_interrupted.
  • Qualified browser reconnect, restart, approvals, artifacts, cancellation, handoffs, workflows, CLI scaffolding, and optional Databricks imports.

0.1.0

  • Established the AgentMuru Python runtime, Workspace, CLI, provider-neutral models, typed tools, permissions, sessions, events, cancellation, approvals, artifacts, traces, workflows, and handoffs.