v0.12.0

0.12.0 (2026-10-03)

Features

• adapter: add bidirectional Rig anti-corruption layer message bridge (877a1d6)
• adapter: decouple sessions, pruning, and providers via Rig anti-corruption layer (8156d10)
• adapter: map tool execution failures to typed Rig ToolErrorKind variants (90d2488)
• engine: migrate to rig 0.43.0 driver and lifecycle architecture (d1e869a)
• mcp: modernize adapter with lazy lifecycle, idle reaper, disk cache, and definition trust (0b53059)
• mcp: support immediate deferred execution and deduplicate ui ansi parsing (962c7a4)
• model: add framework-independent chat message and tool domain types (b673722)
• permission: add critical danger pre-flight intercept to guard evaluator (202886c)

Bug Fixes

• antigravity: format observation text with tool call signature (fc98210)
• antigravity: pass clean response text for unsigned tool observations (7f30cc7)
• engine: configure invalid tool call retry budget on agent runner (cb5a2c6)
• engine: strip canonical model prefix on wire calls and isolate provider extras (d169a16)

Performance Improvements

• core: optimize turn pruning and summary string sanitization (bec9949)
• presentation: optimize kv string sanitization and early-break object formatting (369b1ee)
• tokens: avoid parsing and cloning single-line json arguments in tool call estimation (adc8ef0)
• ui: avoid temporary string formatting allocations in render formatters (f5861d2)

v0.11.0

0.11.0 (2026-09-28)

Features

• cli: support unified model spec in cli flags and env vars (7205624)
• collab: add interactive TUI and modal tool approvals for guest client (66ec257)
• collab: add slash commands, peer manager modal, and status indicators (bdf8733)
• collab: exchange hostname in handshake and track peer identity (1d5af53)
• collab: implement guest client for rho join (c4a6fef)
• collab: implement host collab server and peer multiplexing (dc7182b)
• collab: implement wire protocol and cryptographic handshake (1d4b0f3)
• collab: synchronize footer metrics and session activity to peers (0265f0d)
• collab: synchronize prompt and turn start across peers (c3308b6)
• collab: synchronize shared tool approvals and multiplex host events (2ec3b8b)
• collab: wire /collab command autocomplete and initial snapshot hydration (3c53231)
• compaction: sanitize large command scripts and tool aliases in serialization (3678dd9)
• config: add legacy model config migration and canonical model spec (78f6813)
• config: introduce ModelsConfig role table and enforce serialization hygiene (7ee71ca)
• context: implement cache-aware pruning, artifact spilling, and speculative compaction (5815f3e)
• context: make historical tool pruning adaptive to model context window (16144ec)
• context: wire active model to historical tool pruning and dynamic read padding (0bc0afc)
• context: wire default artifact demotion hook and document optimization spec/plan (b6dc1f2)
• engine: prune historical failed bash commands outside protection window (c9284cc)
• engine: run speculative auto-compaction asynchronously in background (8f74a9e)
• model: implement model spec parser and guard non-thinking invariant (7d097a3)
• permission: add critical danger regex and guard system prompt (3c0f472)
• permission: add guard evaluator and structured verdict parser (ff4fecf)
• permission: align guard prompt with skill builder xml and generic security boundaries (2829371)
• permission: integrate guard model evaluator into permission hook (64b79ca)
• permission: refine prompt notice order, 10s timeout, and built-in baselines (a03f83a)
• repl: support canonical model spec in engine factory and repl model selector (d72c537)
• settings: add guard model configuration and selector (0818dee)
• tokens: calibrate text token estimation for claude models (c876518)
• tools: collapse repeated lines and window clean bash outputs (a9f980c)
• tools: delegate external and protected path gating to permission hook (3968f20)
• tools: group rg matches by file and reduce search limits (87fbb00)
• tools: spill truncated read, rg, and fd outputs to artifact files (116a59e)

Bug Fixes

• compaction: preserve cache prefix and turn boundary during mid-run compaction (eee0ddd)

Performance Improvements

• test: accelerate test suite and coverage execution with nextest (374aaef)

Miscellaneous Chores

• release: trigger 0.11.0 release (3a1eb2e)

v0.10.1

0.10.1 (2026-09-25)

Features

• crap: metric (#50) (f085381)
• exa: gemini search (#51) (f01bcc2)
• hooks: add rtk token optimization hook example and docs (#44) (c539852)
• multimodal analysis (#53) (cb27138)
• tools: add web tool toggles, search engine selector, and settin… (#49) (519fc6f)
• web: add specialized fetch extractors for GitHub and YouTube (#52) (fd907bc)

Bug Fixes

• antigravity: restore daily endpoint priority and non-blocking metadat a failover (#48) (a60127f)
• ci: pages deployment fix remove wasm (6b3ee3a)

v0.10.0

0.10.0 (2026-09-22)

⚠ BREAKING CHANGES

• prompts: Prompt templates in .rho/prompts/ are no longer discovered. Move prompt templates to .agents/prompts/ or prompts/.
• mcp: Project MCP configs in .rho/mcp.json are no longer loaded. Move project MCP configs to .agents/mcp.json or .mcp.json.

Features

• mcp: resolve project mcp config from .agents/mcp.json instead of .rho/mcp.json (f495775)
• permission: support rtk command wrapper in bash analysis and read-only evaluation (50a1704)
• prompts: resolve prompt templates from .agents/prompts instead of .rho/prompts (118711a)

Bug Fixes

• engine: do not prune assistant tool-call arguments during context replay (de3d7aa)

v0.9.0

0.9.0 (2026-09-21)

⚠ BREAKING CHANGES

• hooks: Lifecycle hooks in .rho/hooks/ are no longer discovered. Move lifecycle hooks to .agents/hooks/.

Features

• engine: add transient network error retry across all providers (f0a3cc5)
• hooks: resolve lifecycle hooks from .agents/hooks instead of .rho/hooks (57bce7d)

Bug Fixes

• engine: add antigravity transport error failover and production endpoint priority (f1eb2c7)

v0.8.3

0.8.3 (2026-09-20)

Features

• auth: support non-tty and key-stdin login for scriptability (0d25210)
• build: add windows icon resource and synchronize config docs (d403179)
• cli: wire models subcommand to model store and dynamic discovery (f520128)
• config: support per-provider models table and switch resolution (f494a46)
• repl: make skill discovery, line editor init, and clipboard pasting non-blocking (3aed51a)

Bug Fixes

• auth: strip BOM and fail closed on auth file parse errors (293a67c)

v0.8.2

0.8.2 (2026-09-20)

Features

• quota: add cache invalidation and 30-second cooldown to QuotaTracker (cc898f7)
• quota: refresh quota on turn completion and periodically during active turns (736ef09)
• tokens: account for image tool results in token estimation and memo hashing (fa20189)
• tokens: add model-aware BPE dispatch and image token accounting (588a284)
• tokens: fix compaction accounting, capture summarization cost, and reconcile anchors (5ef0371)
• tokens: optimize message hashing and enforce bounded cache capacity (ffb12e5)
• tokens: prune tool result image blocks from historical conversation context (7f268f4)
• ui: default to native hardware cursor with configurable software fallback (3381378)

Bug Fixes

• engine: allow starting session without network connection (7a09437)
• quota: remove periodic quota polling during idle in REPL and RPC daemon (bdb4530)
• quota: update quotas in real time during idle and turns (9664527)

v0.8.1

0.8.1 (2026-09-19)

Features

• compaction: prune transcript tool results and retry on context overflow (4e39143)
• engine: decouple volatile git status from static system prompt cache prefix (a45ffe0)
• engine: extend historical replay pruning to web, edit, and mcp tools (20723bd)
• engine: implement 4-breakpoint prompt caching and deterministic tool sorting (a9d49b4)
• engine: prune historical read, search, and write replay payloads (69a26ab)
• mcp: implement deferred tool search and dynamic tool activation (df758b1)
• metrics: compute and display prompt cache efficiency and hit ratio (d7f968d)
• repl: display prompt cache hit ratio in session diagnostics (244006c)
• skills: discover project skills from .agents/skills instead of .rho/skills (2ad122f)
• tokens: memoize message token counting and compact tool call arguments (4dce985)

Bug Fixes

• engine: disarm spinner instead of clearing on approval start (048536d)
• repl: flush UI batch immediately on flushing turn events (bfd3797)
• repl: prevent input lockups during tool runs and block streaming (c2098db)
• ui: wrap long words without exceeding block width or dropping spaces (711fe1a)

Performance Improvements

• repl: guard ui channel select loops and offload image processing (1ab61dd)
• tui: atomicize transcript card write and eliminate multi-pass repaints (16fa133)
• ui: lazily read file at most once across edits in diff cards (819c6e0)
• ui: pre-warm syntax highlighter on background thread at startup (ee347eb)

v0.8.0

0.8.0 (2026-09-18)

Features

• compaction: add CompactionSummaryPayload schema and renderer (cf95d6b)
• compactor: integrate DemotionHook for context eviction (e3d3a92)
• compactor: structured summary extraction with multi-stage fallback (0ba9851)
• engine: adopt rig 0.42 model listing, 429 retries, and hook scratchpads (de7afcf)
• engine: adopt rig 0.42 truncation recovery, invalid tool healing, and tool simplification (a9ab82b)
• engine: implement local embeddings, passive RAG, and settings toggle (a37e577)
• engine: prune verbose historical bash outputs across turns (fa7fbc6)
• tokens: implement rig TokenCounter for BpeTokenCounter (4e8d672)

Bug Fixes

• antigravity: align thinking budget and model enum wire protocol with pi-antigravity (47542b1)
• build: allow linker_messages to suppress macOS eh_frame warning (dd30e54)
• chatgpt: track reasoning part coordinates across output items in sse parser (3d2477f)
• provider: restore ollama real context discovery and power-of-two formatting (a75c6b8)
• quota: refresh quota synchronously on model switch and retry with token refresh (3617932)
• ui: eliminate jumpiness, blocking redraws, and spinner starvation (0d2e075)

Performance Improvements

• build: feature-gate fastembed and tune release codegen to reduce binary bloat (30cc80d)
• build: optimize compile profiles and consolidate integration test binaries (52b4b74)

v0.7.5

0.7.5 (2026-09-14)

Features

• ui: support configurable hardware and software cursor modes (c4a4fd5)

Bug Fixes

• antigravity: cap gemini max output tokens at 65535 and extract field violations (eb72e59)
• antigravity: use thinkingLevel instead of thinkingBudget for gemini-3.5-flash-lite (3a3c7ab)
• chatgpt: inject paragraph breaks between reasoning summary parts and unify sse streaming (a971518)
• chatgpt: set store to false and request encrypted reasoning in codex requests (9f5247f)
• claude: track prompt cache tokens and fix request cache breakpoints (96a6d34)
• engine: normalize cached prompt tokens and anchor in-flight context estimation (328311a)
• release: align rho-wasm workspace version (5206eb8)
• ui: sanitize carriage returns in tool output and block rendering (43f8d18)
• ui: strip carriage returns in streaming thinking and verify natural line breaks (232396a)

Performance Improvements

• build: enable fat lto and single codegen unit for release profile (f6b50cc)