Compatibility & Support
What ARI runs on. For the policy around versions (SemVer, support windows, deprecation), see Release & versioning policy.
Python
| Version | |
|---|---|
| Hard requirement | Python ≥ 3.9 (requires-python in ari-core/pyproject.toml) |
| Recommended | 3.10+ (the Quickstart targets 3.10 or later) |
setup.sh checks the interpreter and installs the rest. Run it as your normal user — never with sudo.
Operating systems
| OS | Status |
|---|---|
| Linux | Supported |
| macOS | Supported |
| Windows | Via WSL2 |
Memory backend (Letta)
ARI's memory is backed by Letta (formerly MemGPT) since v0.6.0. setup.sh bootstraps it, auto-detecting the best deployment: Docker → Singularity/Apptainer → pip (skip with SKIP_LETTA_SETUP=1).
The live behaviour is verified against Letta 0.16.7 (see the implementation note in Memory architecture). Check a running backend with ari memory health. Each checkpoint also carries a memory_backup.jsonl.gz snapshot, so a run stays portable even across Letta versions.
LLM backends
Model routing goes through LiteLLM, so any OpenAI-compatible provider works. Select with ARI_BACKEND / ARI_MODEL (always use the provider prefix, e.g. openai/gpt-4o).
| Backend | ARI_BACKEND | Notes |
|---|---|---|
| Ollama | ollama | Local, free, no API key (default for getting started) |
| OpenAI | openai | Cloud, paid; OPENAI_API_KEY |
| Anthropic | claude | Cloud, paid; ANTHROPIC_API_KEY |
| Any OpenAI-compatible | (custom) | Routed via LiteLLM |
Per-phase model overrides are available (e.g. a cheaper model for idea generation, a stronger one for paper writing) — see Configuration and Environment variables.
Skills vs core
Skills are versioned independently of ari-core. A skill at 0.7.x works with any ari-core 0.7.y (compatibility within a minor); across minors, expect a coordinated release. See Release policy → Compatibility windows.
See also: Release policy · About · Quickstart · Environment variables