v2.3.0

Full Changelog: v2.2.0...v2.3.0

v2.2.0 — Native Anthropic backend: route Claude alongside your local models

Albatross now speaks to Claude directly. Add ANTHROPIC_API_KEY, run /backend anthropic, and Claude models join your routing stack alongside local backends — with the same transparent, auditable routing receipts as everything else.

Added

• Native anthropic backend using the documented Claude Messages and Models APIs with ANTHROPIC_API_KEY. (Claude.ai OAuth and subscription tokens are intentionally unsupported.)
• Full agent-loop support: streaming text, tool calls, image input, and thinking-signature round-tripping.
• Honest accounting: cache-read and cache-write token accounting, model/effort receipts, and date-aware Claude Sonnet 5 pricing on the per-turn cost line.

Pair it with routing policies from v2.1.0: put Sonnet 5 on your high tier, a local qwen2.5-coder on low, and let /route report show which model earns its cost.

Install

brew install morganlinton/tap/albatross
# or
cargo install albatross-cli

Notarized macOS binaries are attached below. Site: https://albatross.sh · crates.io: https://crates.io/crates/albatross-cli

Full Changelog: v2.1.0...v2.2.0

v2.1.0 — Transparent routing policies, scoreboards, and outcomes

Transparent multi-model routing gets its audit layer: see every candidate model, why it was or wasn't chosen, what it was expected to cost, and how it actually turned out.

Added

• Routing policies for quality, cost, confidence, local-only operation, effort compatibility, per-turn cost caps, and unknown-cost handling (modelSystem.policy).
• Selector confidence and per-candidate scoreboards with cost estimates, eligibility warnings, policy exclusions, policy hashes, and deterministic fallback explanations in durable routing receipts.
• /route simulate — preview selection without switching the active model.
• /route why-not [model] — inspect candidate eligibility, cost estimates, and policy exclusions.
• /route label pass|fail — record an outcome for the latest route; automatic test results also label the active route.
• /route report — aggregate routing quality, confidence, outcomes, and cost.

Install

brew install morganlinton/tap/albatross
# or
cargo install albatross-cli

Notarized macOS binaries are attached below. Site: https://albatross.sh · crates.io: https://crates.io/crates/albatross-cli

Full Changelog: v2.0.3...v2.1.0

v2.0.3

Full Changelog: v2.0.2...v2.0.3

v2.0.2

Full Changelog: v2.0.1...v2.0.2

v2.0.1

Full Changelog: v2.0.0...v2.0.1

v2.0.0

What's Changed

• fix(grok): send CLI proxy client version and User-Agent by @nelopuchades in #31
• fix: default /login and /logout to the active OAuth backend by @nelopuchades in #32
• perf(agent): run web_fetch in parallel with other read-only tools by @santhreal in #29
• perf(checkpoint): async-read file baselines before mutating tools by @santhreal in #30
• perf(auto): skip done-check until rubric eval passes by @santhreal in #28
• Rename Small Harness to Albatross by @morganlinton in #33

Full Changelog: v1.2.7...v2.0.0

v1.2.7

What's Changed

• perf(cache): keep the prompt-cache prefix stable and request/report the cache by @santhreal in #20

Full Changelog: v1.2.6...v1.2.7

v1.2.6

What's Changed

• feat: arrow-key pickers for setup, /backend, and /model by @nelopuchades in #23
• fix: truncate batch-edit previews on Unicode char boundaries by @santhreal in #24
• fix: cap ship status lines on Unicode char boundaries by @santhreal in #25
• fix: mask API keys by Unicode scalars by @santhreal in #26
• fix: build repo-search snippets without lowercase byte offsets by @santhreal in #27
• feat: add Grok SuperGrok / X Premium+ OAuth backend by @nelopuchades in #18

Full Changelog: v1.2.5...v1.2.6

v1.2.5

What's Changed

• feat: pin backend/model defaults with --default and default-aware pickers by @nelopuchades in #21

Full Changelog: v1.2.4...v1.2.5