akitaonrails/ai-memory v2.1.0

on GitHub

ai-memory 2.1.0

The 2.1 feature train, on top of every 2.0.x fix.

Highlights

Ordered LLM provider fallback chains (#648). Configure [[llm_fallbacks]] in
config.toml with additional providers (each with its own model, optional
base_url, and api_key_env). On a transient failure — 429, 5xx, timeout, or
connection error — the chain advances to the next provider, preserving the
original request, schema, and operation id; deterministic failures (bad request,
auth, unsupported schema) stop immediately. Per-candidate circuit breaker,
resolved-once credentials that never inherit the primary's env var, and passive
per-candidate health in ai-memory status — labels and status only, never keys
or response bodies.

bootstrap --resume (#621, #635). An interrupted bootstrap resumes from
durable per-chunk progress instead of re-paying for every LLM call. It adopts
only the contiguous completed prefix and re-runs from the first gap, so a resumed
run can never diverge from a clean one. Complements the transient-retry from
2.0.3.

First-party OpenCode 2.0 beta (#622). install-mcp --client opencode2,
install-hooks --agent opencode2, and ai-memory run opencode2 — full lifecycle
capture, handoff injection, and managed launch/resume/import against the V2
plugin shape. Shares v1's config dir and session store; no store migration.

Codex native SessionEnd (#604, #605). install-hooks --agent codex now wires
Codex's first-class SessionEnd event (Codex CLI 0.145.0+), so a finished Codex
session gets the same automatic end-of-session summary and cross-agent handoff
Claude Code already gets.

Operator LLM headers (#606). AI_MEMORY_LLM_HEADERS / llm_headers attaches
caller-identifying HTTP headers to every LLM request, validated once at the config
boundary and marked sensitive on the wire.

Also in 2.1.0 (fixes forward-merged from the 2.0.x line)

• Auth-disabled servers stop returning 401 for a stale bearer (#639); invalid
  bearers still rejected whenever any auth mode is enabled.
• Cursor sessions are filed under their repo and the cursor agent instead of
  default/scratch + claude-code (#651).
• reindex no longer aborts on a scope created after startup — the manifest is
  written with the scope's first page, and a missing manifest names its path
  instead of a bare os error 2 (#643).
• purge-session removes the on-disk wiki page so a later reindex can't
  resurrect a forgotten session (#653).
• [briefing] max_chars now bounds the whole brief, never just the bodies, and
  the mandatory untrusted-history scaffold is always preserved (#657).
• Static-vs-session MCP client routing guidance prevents another session's active
  project from capturing an unscoped write (#372).
• Podman-without-Docker support in the bin/ai-memory wrapper (#636).
• Self-contained build (vendored Tailwind default, no build-time network) and a
  faster two-tier dev test loop (#645).

See CHANGELOG.md
for the full list.

Checksums (SHA256)

815fdb3ff32d630aaf7778af7bc01b18ed6fb3fdf2d43de28de57b21bb46312b  ai-memory-hooks.tar.gz
60cdc739c21f38b017c89f7fdc5708eb8d08fec715dfa62f364c23698094b074  ai-memory-install-hooks
3eafda7afaba63008c2b0e27713c1076dff3b366c377c30f43a7b04a0d78b6c1  ai-memory-linux-aarch64.tar.gz
b4d911f392fea7f151c9c1ff83521922417570e5104ffb02a5e82253afc877ac  ai-memory-linux-x86_64.tar.gz
61811954da0ee1519024f81271cb7b34893a28d7c9e3d14d07eb80fadcf44d58  ai-memory-macos-aarch64.tar.gz
13259e2d60705d09b8b7dd8bbcea0d361b994498333095c91d0128f382a6a1da  ai-memory-macos-x86_64.tar.gz
72c477e5c02b54c202e1c7515aacf00a1702f9916c51a26d875fcfeff65b1451  ai-memory-windows-x86_64.zip
6a91c44ffa2e85d3b5a6ce26a4ccff0286b91d08d276d2bb4d47c346b036ee8b  ai-memory-wrapper.cmd
2d35c2f1fba28a1aee3ffeb58e3deaec772803f3dc6fd159de6ae0a47ba797ac  ai-memory-wrapper.ps1
127282baf003aec6a5e3a8c73c5186ff785c2a412c93e5538141ceb9b04909d9  ai-memory-wrapper

Install

# Arch Linux (AUR)
yay -S ai-memory-bin      # prebuilt Linux x86_64/aarch64 binary
yay -S ai-memory          # builds from source

# Docker
docker pull akitaonrails/ai-memory:2.1.0

# macOS aarch64/x86_64 (no toolchain): download the matching ai-memory-macos-<arch>.tar.gz below,
# extract it, and follow the bundled docs/macos.md.

# Windows x86_64 (no toolchain): download ai-memory-windows-x86_64.zip below,
# extract it, and follow the bundled docs/windows.md.

# From source
cargo install --locked --git https://github.com/akitaonrails/ai-memory --tag v2.1.0