Skip to content

CHANGELOG.md

Latest commit

 

History

History
632 lines (531 loc) · 39.2 KB

CHANGELOG.md

File metadata and controls

632 lines (531 loc) · 39.2 KB
Raw

Changelog

Unreleased

0.2.0 - Unreleased

Added

  • Added /transcript to browse long TUI sessions without depending on terminal scrollback, with line, page, and first/last navigation.

Fixed

  • Fixed a renderer crash dialog reporting React error #185 ("Maximum update depth exceeded") coming from the composer's prompt-history inline completion (#4117): the offer engine the 0.1.11 composer fed could flip-flop its announcement state on real-layout measurements until React hit its nested-update limit, which surfaced as the crash dialog. The unstable completion wiring was removed from the composer (#3292), Astryx 0.5.0 no longer ships the engine (#3755), and regression tests now keep that seam closed.

Changed

  • Made typed request() the sole direct Runtime Host operation API; removed the 17 forwarding aliases from direct and reconnecting connections while preserving status validation, subscriptions, capabilities, listeners, lifecycle, and close behavior.
  • Collapsed the RuntimeRunner/Flow/Invocation shell into RuntimeKernel; backend dispatch, terminal coalescing, stop/drain, and durable continuation admission now have one production owner, immutable request snapshots remain enforced at AgentRun acceptance and backend dispatch, and SessionEvent-to-RuntimeEvent conversion remains a pure mapper.
  • Retired the Task Ledger domain: SessionTodo is now the sole authority for in-session work items, and the operational-state schema drops the workflow_task_ledger_events table on first open. Unfinished Tasks are not migrated and are permanently deleted. This affects workspaces last opened by v0.1.0 through v0.1.11, cli-v0.1.0-beta.1, v0.2.0-incubating-rc1, or a v0.2.0-dev build; those releases wrote Tasks to a table that no shipped build ever bridged into SessionTodo. Before opening such a workspace with this build, finish or export the Tasks you still need, or copy the workspace's runtime.sqlite aside — the migration removes the only live copy, so afterwards recovery requires a backup made in advance.
  • A compaction rejected as too large for the summarizer's own window now retreats to the span the last accepted request's input covered, instead of halving the covered range. That span is the newest reply this route produced, found through the run headers, so it was accepted by this model on this connection and is provably within capacity; halving can overshoot (discarding verbatim history for nothing) or undershoot (paying another round trip), and a span another route accepted proves nothing at all. One retreat, then the fold fails open and the provider decides.
  • token_usage anchors now record the model and connection that produced them. A token count is a number in one model's tokenizer against one connection; carrying the route on the record lets any reader apply the rule the runtime already enforces, instead of pairing one model's usage with another model's window. The record decodes against a closed allowlist, so sessions written with these keys do not open in earlier releases, and the Runtime Host compatibility epoch moves to 107.
  • The provider-dropping note now also fires across the send boundary. A provider that truncates to a fixed window reports the same input on every later request while the user keeps adding turns, which a send of one or two steps cannot see from the inside; the first request of a send compares against the persisted anchor instead. Across the boundary the test is equality rather than "did not grow": inside a send Maka knows it only appended, while across it a manual compaction, a smaller tool set or an edited history all shrink the input legitimately, and none of them lands on exactly the same count. The note carries the two counts it compared, and is reported once per backend rather than once per send, because the condition persists once it starts.
  • Let the provider decide whether a request fits. Proactive compaction now uses only a user-declared Maka window and the previous accepted request's provider-reported inputTokens + outputTokens; no declaration means no proactive capacity threshold. /models and generated model metadata are display hints, not limits. token_usage records persist the last-request anchor under lastRequestAnchor; its new { inputTokens, outputTokens } shape still decodes the retired payloadChars key from older sessions. Requests that are too large are compacted and retried once after a real provider rejection, then reported as a context_overflow provider error. Compaction is entered at most once per send, and a request rejected after a fold was actually applied is reported as still too large after compaction. A fold that failed open makes no such claim: that request went out with its full raw history. A reply cut at finishReason: length no longer triggers a fold, because the provider running out of window room and the provider's own lower output cap are indistinguishable from outside. Five system notes explain the provider-side cases: dropping context, a window worth declaring, an exchange past the declared window, a request accepted past the window the model reports (once per crossing, while nothing is declared), and a request still too large after compaction. The reply reserve that arms the proactive threshold is twice the last real reply, bounded at 8,000 tokens, rather than the model's maximum output. Sessions this build writes do not open in earlier releases: those decode token_usage against a closed allowlist, so the reshaped lastRequestAnchor key fails the record and, with it, the Session that contains it; downgrading therefore needs a copy of the workspace's runtime.sqlite taken before the upgrade. Nothing produces the context_budget_exhausted stop reason any more — a request that really is too large is compacted and retried once, then reported as a context_overflow provider error — though sessions that already recorded it still decode and present. The Runtime Host compatibility epoch moves to 106.
  • Unified context management under one Runtime-owned policy. MAKA_CONTEXT_* environment overrides no longer tune or disable compaction and Tool Result pruning; model-visible archive placeholders are read on demand through bounded ArchiveRead calls instead of eager hydration. Previously supported overrides are ignored on upgrade: if Tool Result pruning was set to off, pruning is re-enabled, and there is currently no supported replacement opt-out.
  • Moved Read image snapshots into the durable context-offload store with Runtime-owned lifecycle identity, exact branch and revision copying, recovery-safe cleanup, and bounded physical garbage collection after Session retirement.

0.1.11 - 2026-08-18

Highlights

  • Expanded Runtime Host from a local execution service into the shared authority for multiple connected Hosts, remote project registration, live run state, and archived session lifecycle (#3097, #3145, #3079, #3074, #3151).
  • Added the installable Maka CLI package and its protected staged npm release pipeline, including cross-platform artifact and Eval validation (#3169, #3173, #3185, #3188, #3192, #3197, #3200, #3201).
  • Added brokered Windows AppContainer sandbox support and tightened local IPC ownership and ACL enforcement (#2961, #3179, #3182).
  • Added Work Board storage foundations, Host-scoped task creation, prompt-history completion, and first-run viewport containment (#3028, #3122, #1874, #3195).
  • Added native Desktop and TUI locale authorities and Qwen3.8 Max Token Plan support (#2686, #2691, #3157).

Reliability and developer experience

  • Preserved live turns across refresh and projected authoritative live execution state through Runtime Host (#3189, #3079).
  • Kept restored tasks safe from concurrent removal, queued busy-raced sends as steering, bounded summarizer inputs, and retired obsolete compact and compatibility paths (#3056, #3032, #3113, #3128, #2742).
  • Hardened MCP rediscovery, provider failure diagnostics, rate-limit handling, session stream completion, and scheduled-task ownership (#2989, #2675, #3115, #2682, #2655).
  • Added fair multi-arm Eval infrastructure and the DeepSeek Harness benchmark arm, while isolating subject metering from framework accounting (#2668, #2971, #3176).
  • Strengthened Windows installer, crash-recovery, remote service restart, and release artifact coverage (#2650, #2562, #3186, #2941).

Fixed

  • Disabled TUI taskbar-progress keepalives by default on native Windows and Windows Terminal sessions, where repeated OSC 9;4 updates can make Explorer's taskbar unresponsive. MAKA_TASKBAR_PROGRESS=1 restores the prior behavior, while MAKA_TASKBAR_PROGRESS=0 disables it explicitly.

Distribution

  • Ships for Apple Silicon macOS as a signed and notarized DMG and ZIP, and for Windows x64 as an unsigned NSIS installer and ZIP, built and verified in the same release run.
  • The bundled Computer Use skill ships with the app, but the Computer Use executor remains excluded from this release.

0.1.10 - 2026-08-10

Highlights

  • Rebuilt the Runtime Host around durable ownership and recovery: clients now reconnect across host restarts, incompatible host epochs retire cleanly, and evicted session streams recover without losing the active conversation (#2613, #2618, #2630, #2633).
  • Added remote capability provider mode and stopped agent graph supervisors from generating unnecessary wake-ups (#2625, #2626).
  • Made the changes panel Git-authoritative, restored slash command discovery, exposed archived conversations, and added an About-page update check (#2610, #2612, #2573, #2629).
  • Added managed dependency artifact authority to storage and routed DeepSeek V4 Flash edits through ApplyPatch (#2485, #2606).

Reliability and developer experience

  • Recovered stuck xAI login attempts and stale connection deletion, kept settings-controlled session chrome hidden, rendered subagents as compact rows, and aligned the workbar picker with Astryx (#2615, #2617, #2619, #2632).
  • Removed obsolete Runtime Host transition residue and preserved the storage hydration crash fixture (#2635, #2576).
  • Reduced Windows CI work by scoping the baseline to affected surfaces and gating expensive storage coverage, while retrying transient Electron downloads (#2599, #2637, #2594).
  • Updated bundled Computer Use and WebContent dependency pins, including the frame reflow fix (#2627, #2631, #2638).

Distribution

  • Ships for Apple Silicon macOS as a signed and notarized DMG and ZIP, and for Windows x64 as an unsigned NSIS installer and ZIP, built and verified in the same release run.
  • The bundled Computer Use skill ships with the app, but the Computer Use executor remains excluded from this release.

0.1.9 - 2026-08-09

Highlights

  • Completed the Runtime Host M5 production cutover (#2420), then established the local standalone service (#2583), authenticated WebSocket access (#2591), and Host-owned project catalog authority (#2603).
  • Added external session import end to end: the shared foundation (#2500), a Codex session adapter (#2502), and the Desktop import flow (#2507).
  • Shipped Visual System 2.0 from its foundations through the full theme sweep and detail polish (#2525, #2536, #2538), aligned high-traffic chrome with Astryx primitives (#2580), and closed the remaining elevation, rhythm, and primitive review debt (#2593).
  • Made task submission readiness a shared product contract (#2498), exposed it in Desktop (#2519), added CLI preflight (#2524), and consumed the readiness result at submission time (#2523).
  • Expanded implementation-agent capabilities with portable terminal input (#2526), semantic terminal mouse input (#2533), interactive shell controls (#2561), and OpenAI native ApplyPatch (#2532).
  • Replaced the CodeMode Self evaluator with QuickJS (#2549).

Reliability and developer experience

  • Runtime recovery now handles idle and incomplete provider streams (#2535, #2604), preserves compaction projection after overflow (#2602), retains imported session context (#2579), and keeps full access available in Plan mode (#2581).
  • Stabilized Runtime Host session lifecycle races (#2548), added client request backpressure (#2539), made task refusals actionable (#2527), recovered CLI sessions after workspace moves (#2531), and kept renamed Claude model ids stable across catalog refreshes (#2482).
  • Matured the generated-files workbar (#2506), kept generated files under user control (#2585), restored composer drafts across remounts (#2584), exposed diagnostics from error toasts (#2540), and added per-connection model request customization (#2565).
  • Cut CI wall-clock time with impact gates and a single end-to-end job (#2589), reduced Storybook to render smoke (#2582), and expanded Windows process, named-pipe, artifact, workspace, and crash-recovery coverage.

Distribution

  • Ships for Apple Silicon macOS as a signed and notarized DMG and ZIP, and for Windows x64 as an unsigned NSIS installer and ZIP, built and verified in the same release run.
  • The bundled Computer Use skill ships with the app, but the Computer Use executor remains excluded from this release.

0.1.8 - 2026-08-08

Highlights

  • Added Codex-style side conversations: branch a side thread from the one you are in and come back with the answer (#2428).
  • Gave projects a first-class Settings page: manage every project Maka knows, pick the default one new conversations open in (explicit default beats last-used; the composer can still switch per conversation) (#2446), rename and reveal a project from its row menu (#2447), and carried the list on the entity-list components with folder anchors and tail-preserving path truncation (#2451).
  • Closed three self-serve gaps reported by users: the Models page can set the default connection where the 默认 badge lives (#2421), new conversations take a configurable default thinking level and the composer menu says 模型默认 instead of a deceptive 默认 (#2430), and the 外观 page structure, grouping, and copy were brought under the settings idiom (#2343).
  • Introduced custom pets end to end: a pack contract with atomic storage (#2422), safe pack import (#2423), a pack library (#2427), persisted selection (#2437), management in Settings (#2440), rendering (#2442), and runtime-state-driven animation (#2444).
  • Enabled Mimo and DeepSeek free models by default for OpenCode Free (#2431) and seeded its default inventory on create (#2443).
  • Removed the product Voice module end to end (#2426).
  • Closed the Runtime Host M4 readiness gaps (#2419).

Reliability and developer experience

  • Windows: serialized root marker repair (#2438), stabilized portable path tests (#2395), enabled long paths for bare Git fixtures (#2435), normalized CI process identities (#2334), and refreshed the test skip inventory (#2434).
  • Storage: removed worktrees outside their cwd (#2424).
  • Renamed the desktop UI components to kebab-case (#2417), finished two CI cleanups and reformatted what landed unformatted (#2432), and trimmed low-value guard and CSS-contract suites while fixing sqlite warning noise (#2425).

Distribution

  • Ships for Apple Silicon macOS as a signed and notarized DMG and ZIP, and for Windows x64 as an unsigned NSIS installer and ZIP, built and verified in the same release run.
  • The bundled Computer Use skill ships with the app, but the Computer Use executor remains excluded from this release.

0.1.7 - 2026-08-07

Highlights

  • Rebuilt the composer attachment experience: staged attachments are Token chips with tooltip metadata and a lightbox (#2393), with preview thumbnails and file cards (#2367).
  • Unified the app's status language: what a status means is now named in one place (#2397), the Settings surfaces state semantics instead of colours (#2401), and the migration finished by deleting its own bridge (#2403) — every status dot draws from a single definition, and five formerly "active-blue" informational dots settled into their true quiet states.
  • Added bounded long-term memory extraction (#2117) and a runtime WebFetch tool (#2362).
  • Moved the CLI onto Runtime Host: maka run (#2337) and TUI sessions (#2308), and scanned finished runs' trajectories for retrieval (#2319).
  • Kept a running status line up for the whole live turn (#2356), opened linked subagents from stream tools instead of the sidebar (#2383), restored the transcript's markdown rhythm (#2348), handed the chat meta row back to Astryx primitives (#2358), and exposed recovery for empty failed turns (#2381).
  • Rebuilt the extension module pages on the shared ModulePage shell (#2266), led the session trace with an Astryx-native overview (#2289), and unified the workspace picker onto the composer's ghost-menu family (#2287).
  • The window titlebar states the session's project and name (#2327), and the Pinned and Recent group headers carry a new-task trigger (#2364).
  • Upgraded Astryx 0.2.0 → 0.3.0 (#2288), unified lucide-react onto one major (#2275), added a dependency audit lane (#2223), and published the bundled Git runtime's source materials (#2235).

Reliability and developer experience

  • Provider transport: hardened incremental OpenAI Responses (#2247), stopped reporting a truncated provider stream as a finished turn (#2297), sent the Responses wire the options it reads and read the reasoning it returns (#2328), settled stop cleanly when it aborts a pending question (#2257), and bounded oversized prior turns (#2378).
  • Sessions and storage: made session copy retries idempotent (#2398), preserved operational state across schema upgrades (#2361), imported legacy JSONL session transcripts into SQLite (#2263), closed usage stores after lease revocation (#2365), made active session joins replay-safe (#2368), and validated swarm resume by agent id (#2375).
  • Chat and UI details: kept quote layer geometry stable during entry (#2377) and shown only for settled selections (#2350), made the tool diff readable with its counts on a collapsed group (#2300), kept the prompt rail clickable on macOS (#2338), used standard tab order in the provider catalog (#2306), and served Claude OAuth models from the curated catalog with real error causes (#2336).
  • Desktop platform: isolated the dev build's userData root from release (#2292), closed xAI OAuth callback connections and retried a busy port (#2302), enforced a metadata-free renderer startup boundary (#2176), and dropped the consumer-less update-download bridge (#2326) while giving the update action a slot instead of a row (#2320).
  • Runtime and headless: compacted and sanitized agent tool contracts (#2349), classified an expired probe budget as timeout (#2344), decided the tool call event's ledger lane at push time (#2240), separated request latency from liveness (#2392), measured the wire the Maka runtime actually dials (#2286), and mounted build outputs into the task container rather than the repo root (#2298).
  • Trimmed low-value and redundant tests across suites (#2404, #2406), and made ordering assertions independent of fixed wall-clock budgets (#2304).

Distribution

  • Ships for Apple Silicon macOS as a signed and notarized DMG and ZIP, and for Windows x64 as an unsigned NSIS installer and ZIP, built and verified in the same release run.
  • The bundled Computer Use skill ships with the app, but the Computer Use executor remains excluded from this release.

0.1.6 - 2026-08-06

Highlights

  • Extended Runtime Host with a Bot adapter (#2192), candidate composition (#2194), and complete opt-in parity (#2216): restored the Desktop Computer Use capability (#2233) and its presentation parity (#2254), and resolved OAuth tokens for connection effects (#2246).
  • Added restricted CodeMode execution in the runtime (#1648), and made Edit/Write/FormatJson emit file_diff results for both renderers (#2264) so tool output renders consistently across Desktop and CLI.
  • Bundled and verified a hermetic Git runtime for managed workspaces (#2199), and gated managed workspace reads behind owner-bound Runtime Host execution (#2129).
  • Implemented the Session Bundle filesystem codec in storage (#2013).
  • Rebuilt the tool detail panel as one surface with one type scale and real diff colours (#2227), kept it visible during live turns (#2244), kept tool-call groups collapsed until the reader opens them (#2226), and kept composer model/thinking pickers mounted mid-turn, unified as ghost menus (#2230).
  • Pinned the prompt anchor rail to the Astryx chat scrollport (#2161) and gave it motion (#2215), dropped the branch picker and rebuilt the project picker on Selector (#2217), and localized the Astryx chrome adopted since #1795 (#2202).
  • Let the Maka benchmark arm run where its competitors run (#2177), bound managed shell sessions to the in-container tool surface (#2258), and recorded the four-arm Terminal-Bench 2.1 comparison (#2208).
  • Pinned GitHub Actions and configured Dependabot (#2212), added a non-blocking Windows baseline lane (#2173), and bumped the AI SDK while re-porting the tool-call tracker patch (#2193).

Performance

  • Streamed OpenAI Responses incrementally (#2238), coalesced partial stream persistence (#2157), mounted a switched-to transcript progressively (#2191), and seeded transcript geometry before progressive fill (#2237).

Reliability and developer experience

  • Enforced swarm failure boundaries (#2203), kept PTY tool results JSON-safe (#2209), replayed provider-facing MCP arguments (#2243), preserved reasoning across provider steps (#2249), and made the tool-result archive capability indivisible (#2255).
  • Preserved a new task across renderer reloads (#2174), exposed thinking levels for discovered Codex OAuth models (#2250), completed the macOS quit lifecycle (#2251), cleared the "New messages" indicator at the bottom and per conversation (#2211), and replaced the unhealthy opencode-free bootstrap model (#2166).
  • Repaired the five regressions from the 定时任务 rebuild (#2261).
  • Kept modal dialogs out of the window drag region (#2210), drew the reasoning chevron from the Astryx icon registry (#2228), preserved Astryx List accessible names (#2225), stripped ATX heading markers from quote chip display (#2214), dropped the dead [data-trow] chat rules (#2231), landed a switched-to session at its latest turn instead of flying there (#2239), and bounded progressive fill idle so long transcripts settle (#2259).
  • Aligned the Models header and back button (#2198), and aligned the Settings chat-defaults column with a resizable assistant-tone textarea (#2256).
  • Made attachment drop readiness atomic in e2e (#2229), and extracted the shared sanitizeUnicodeText into core (#2190).

Distribution

  • Ships for Apple Silicon macOS as a signed and notarized DMG and ZIP, and for Windows x64 as an unsigned NSIS installer and ZIP, built and verified in the same release run.
  • The bundled Computer Use skill ships with the app, but the Computer Use executor remains excluded from this release.

0.1.5 - 2026-08-05

Highlights

  • Added provider-native web search: one WebSearch routing layer that selects the current model provider or explicit Tavily execution, with OpenAI and Anthropic implementations, durable replay, citation metadata, and privacy-mode coverage across Desktop, CLI, Runtime Host, and opt-in Headless Harbor (#2152).
  • Extended Runtime Host ownership to Plan turn transitions (#2169), Plan runtime state (#2082), Daily Review runtime (#2113), Deep Research state (#2099), and derived Session effects (#2066), and added its Desktop client foundation (#2134): session adapter (#2140), catalog (#2143) and execution (#2149) IPC, session domains (#2164), effective pricing projection (#2073) and its Pricing adapter (#2148), and the native Browser and Computer Use capabilities Desktop offers back to the host (#2178).
  • Instructions now load from ~/.maka/AGENTS.md (or CLAUDE.md/GEMINI.md) before the project's own, so a preference set once applies in every workspace (#2183).
  • Bundled an English Computer Use Skill tailored to the shipping semantic action surface, auto-installed only when a Computer Use backend is available without overwriting modified or untrusted copies (#2147).
  • Made app updates background and task-aware: discovery and download now live in Electron main, and restart/install is guarded by a main-owned activity snapshot that confirms how many tasks it interrupts (#1992).
  • Restored the native application menu and unified platform command routing (#2098).
  • Gated managed-workspace execution behind an owner-bound, short-lived authority scope (#2106) and sealed baseline admission with a durable Git receipt (#1872), so a capability is usable only while the storage root, workspace binding, and canonical head describe the same Git state.
  • Continued the Astryx redesign of Settings: every capability gets its own disclosure (#2100), a settled value stays a row until you ask to edit it (#2139) behind one expandable row component instead of two (#2141), and the specs and jargon nobody can act on are gone (#2118, #2116, #2120); project files left Settings (#2081), the junk drawer emptied into the workbar toggle (#2123), and project and branch joined the composer footer control row (#2168). Usage (#2024) and Skills (#1973) followed, then three more surfaces: 「计划提醒」and「每日回顾」became one dense row-and-inspector page instead of two unrelated implementations (#2155), the session workbar dropped its hand-rolled shell and the tokens that silently resolved to fallbacks (#2158), and the MCP page came down to one primary action (#2035).
  • Let the Inspector trace filter by what a reader actually asks (#2115), and closed the follow-ups left on the trace panel (#2041).

Performance

  • Lazy-mounted sidebar session trees (#2021).
  • Removed the swarm ledger full-scan hot path (#2040).

Reliability and developer experience

  • Fixed copied tool authority facts on branched tools (#2061), preserved stream liveness across filtered parts (#2124), returned proxied responses at headers instead of body EOF (#2159), landed the terminal fact a stop claims (#2078), and made the execution boundary the sole file-path authority (#2087) in one realpath space (#2059).
  • Made realtime session creation single-flight (#2135), released settled bot typing listeners (#2150), honored the default permission mode immediately (#2109), reset incompatible operational state (#2122), resolved permission overlay assets in bundled dev (#2045), and honored TMPDIR for Runtime Host control endpoints (#2160).
  • Bounded Headless teardown and marked the output replay (#2151), reaped the scoped command's replay (#2146), kept Harbor archive ids within ref grammar (#2130), stopped handing graded competitor arms this repo's tree (#2121) and misfiling their cells as infrastructure failures (#2090), let an authoritative reward survive a missing agent self-report (#2112), made Maka's settlement window reachable at the Harbor boundary (#2107, #2114), honored the command-timeout floor (#2110), resolved host.docker.internal for in-container arms (#2062), unified canonicalJson into a single serializer (#2005), pointed the AHE snapshot at the moved workspace-instructions source (#2092), added the Reasonix benchmark arm (#2079), and raised the A/B pair-concurrency cap to 16 (#2075).
  • Rehydrated an unanswered user question the surface never received (#2086), and made thinking strength follow the chosen level on Kimi/StepFun/MiniMax coding plans (#2067).
  • Polished the shell and chat surfaces: an opaque popover for selection quote actions (#2154), a centred empty-chat hero (#2153), traffic lights aligned with the sidebar icon column (#2144), a collapsed rail that stops repainting a third tone in dark mode (#2187), one 920px content column per settings section (#2080), a row hairline that stops bending on Astryx Item's corners (#2111), Astryx Link for the last bare anchors (#2138), a localized required/optional field marker (#2184), global shortcuts through Astryx useHotkeys (#2091), and Astryx clickAction owning the in-flight button state (#2089).
  • Fixed the 定时任务 inspector to sit two tab stops from any row (#2185) and seeded scheduled tasks with distinct createdAt (#2186).
  • Added an a11y audit that flags aria-label on elements whose role cannot hold a name (#2108), and established the Windows support baseline (#2156).
  • Replaced fixed waits with explicit barriers in runtime tests (#2162), owned the temp namespace in the test runners (#2068), isolated the dev bootstrap environment (#2131), seeded a real project in the sidebar rename spec (#2022), eliminated Desktop smoke false failures (#2101), stabilized Skill draft restoration (#2137) and invocable Skill projection (#2105), built an invalid UTF-8 path with Git plumbing (#2097), proved the Storybook autoplay contract the smoke harness stands on (#2071), and fixed the serial workspace batch that never ran after a parallel failure (#2136).

Removed

  • Removed dead storage modules (#2104), consumer-less Desktop IPC bridge surface (#2065), and dead shell CSS recipes along with the check-dead-css blind spots that hid them (#2070).

Distribution

  • Ships for Apple Silicon macOS as a signed and notarized DMG and ZIP.
  • Adds Windows x64: an NSIS installer and a ZIP, built and verified on Windows in the same release run (#2182). This build is not code-signed yet, so Windows SmartScreen warns on first launch.
  • The bundled Computer Use skill ships with the app, but the Computer Use executor remains excluded from this release.

0.1.4 - 2026-08-04

Highlights

  • Rebuilt Computer Use around maka-cu as the executor that actually runs: rewrote what the model reads (observation, action surface, refusals), made the agent cursor land where it is aimed, mirrored the driven window instead of competing for the screen, and let a model read back the call it made.
  • Gave driving the computer its own conversation row, stopped a run when the machine locks, and kept the machine awake while it runs.
  • Made SQLite the sole operational authority in storage, moved the project catalog into the operational database, and converged concurrent fresh WAL initialization.
  • Extended Runtime Host ownership over input and artifact preparation, interactive session actions, safe-boundary continuation authority, and sandbox boundary interactions.
  • Added a session trace as a workbar Inspector tab, projected over both ledgers.
  • Added configurable subagent model routing with its own settings page, and raised agent swarm concurrency to 32 ways.
  • Added Mermaid rendering to Markdown.
  • Switched Codex OAuth to the ChatGPT device-code flow and added Grok PKCE.

Performance

  • Streamed the transcript incrementally instead of re-deriving it per token.
  • Subscribed to session UI state at the granularity each surface reads, and stopped idle-session event-health polling from re-rendering the whole shell.

Reliability and developer experience

  • Fixed turn execution scope isolation, streamed tool-call delta resolution against live aliases, streamed tool calls dropped when a gateway skips index 0, tool-free child summaries, and AgentRun ledgers written by other versions.
  • Made macOS development permission grants reliable and the dev TCC grant durable behind an opt-in, and made startup say which step has not come back.
  • Told a refused tool call what to do next, not just what went wrong.
  • Derived thinking-strength controls from models.dev reasoning options, and unified cron expression authority.
  • Continued the Astryx redesign across Settings, the subagent page, the extensions page, scheduled tasks, skill chips and empty states, and the composer model and voice controls.
  • Put every headless benchmark arm under the same tool surface and model budget, registered ArchiveRead for Harbor-archived tool results, and required validated verifier grades.
  • Routed CI into dedicated lanes for heavy suites and the alignment audit, and measured Computer Use rendering changes against recorded trajectories.

Removed

  • Removed cua-driver now that maka-cu is the Computer Use executor.
  • Removed isolated runtime utilities and the Cursor subscription integration.

Distribution

  • Ships for Apple Silicon macOS as a signed and notarized DMG and ZIP.
  • Computer Use remains excluded from this release.

0.1.3 - 2026-08-03

Highlights

  • Expanded Runtime Host ownership across hosted child agents, Agent Graph execution, Web Research, Goals, Automation, OAuth enrollment, session lifecycle, runtime resources, and live execution inspection.
  • Strengthened Graph and Swarm coordination with structured operator handoffs, non-blocking supervisor turns, durable child summaries, and session-bound graph retirement.
  • Migrated the Desktop shell, Settings, Composer, conversation surfaces, typography, navigation, resizing, and common controls to Astryx 0.2.0.
  • Added SQLite-backed local memory, managed Git workspaces, verified operational backups, canonical model-call metering, and confirmed agent self-configuration tools.
  • Added opencode-free as a zero-credential default provider and improved Runtime Host OAuth-backed execution.

Reliability and developer experience

  • Fixed Electron development and packaged startup failures, onboarding snapshot handoff, Runtime lifecycle races, PTY backpressure, clipboard writes, and several Graph, Headless, and UI regressions.
  • Reduced low-signal, duplicate, source-shape, and happy-path tests; routed CI suites by affected surface; and kept build, typecheck, Storybook, and E2E gates intact.

Removed

  • Removed the desktop Open Gateway HTTP/SSE server and its Settings and IPC surfaces. Model providers, OpenAI-compatible connections, Vercel AI Gateway, and bot event gateways are unchanged.
  • Removed the superseded Expert Team mode and its Notes-era architecture artifacts now that Graph and Swarm own multi-agent coordination.
  • Retired legacy storage writer exports and Astryx compatibility adapters that no longer had active callers.

Distribution

  • Ships for Apple Silicon macOS as a signed and notarized DMG and ZIP.
  • Computer Use remains excluded from this release.

0.1.2 - 2026-07-31

Runtime kernel extraction

This change set turns the runtime execution path from a large implicit SessionManager / AiSdkBackend flow into an internal runtime-kernel shape. It keeps the existing desktop, renderer, IPC, session JSONL, settings, and bot surfaces stable while moving model, tool, trace, run-ledger, and startup-recovery responsibilities behind explicit internal boundaries.

Area Summary
Tool runtime Extracted an internal ToolRuntime around tool input validation, permission checks, watchdog pause/resume, abort propagation, telemetry, artifact recording, and failure classification.
Model adapter Extracted a minimal ModelAdapter so provider stream/error/usage normalization no longer lives directly in the backend orchestration shell.
Runtime trace Added best-effort RunTrace events for model, tool, permission, abort, and usage milestones without changing renderer-visible SessionEvent behavior.
AgentRun ledger Added core AgentRun types and a file-backed AgentRunStore at sessions/<sessionId>/runs/<runId>/run.json plus events.jsonl.
AgentRun execution Moved the heavy turn execution lifecycle from SessionManager.sendMessage() into internal AgentRun.execute(), including user-message append, backend stream drive, status projection, abort/failure handling, and durable trace writes.
Startup recovery Made recoverInterruptedSessions() prefer the AgentRun ledger when available, repairing stale non-terminal runs and preserving the legacy message/turn-state fallback for older sessions.

See docs/archive/runtime-kernel.md for the historical design rationale, boundaries, and verification details.

Hardening phases 1-5

This change set collects the first five maintenance hardening phases from the Rive deep-read follow-up work.

Phase Area Summary
1 Runtime permission and usage handling Made stream watchdog pause/resume accounting robust for concurrent tool calls, added permission timeout handling, integrated Office document abort propagation, and fixed cache/reasoning token usage extraction.
2 Session JSONL recovery Recovered sessions with corrupt JSONL rows by parsing message lines independently, surfacing landed corrupt rows as system_note, and dropping malformed truncated tail rows.
3 Bot and OpenGateway abuse controls Added bot inbound rate and session-binding limits, bounded bot dedupe state, forced bot-bound sessions to explore, and capped OpenGateway SSE connections with idle cleanup.
4 Credential-store secret kind expansion Extended encrypted credential-store support for bot tokens, bot app secrets, proxy passwords, gateway tokens, and Tavily API keys while preserving legacy API-key/OAuth-token key formats.
5 Connection credential IPC input hardening Added shared main-process validation for renderer-controlled connection slugs and API keys before store, credential, or provider side effects.

Active tool-result pruning default-on

activeToolResultPrune (current-turn large tool-result pruning) is now enabled by default on both desktop and headless. Tool results above the 2048 estimated-token threshold are archived and replaced with a metadata-only placeholder (artifact id + content hash) in the next provider-visible request; the raw payload is preserved in the archive and is not lost. On desktop this runs before the already-default-on semanticCompact summary step. On headless the placeholder reaches the next provider step directly (no default compaction/retrieval), which benchmark A/B evidence in #340 showed is non-inferior within the 10pp margin while saving ~31.6% cost.

Opt out with MAKA_CONTEXT_ACTIVE_TOOL_RESULT_PRUNE=off (both sides). Tune the threshold with MAKA_CONTEXT_ACTIVE_TOOL_RESULT_MAX_ESTIMATED_TOKENS and the start step with MAKA_CONTEXT_ACTIVE_TOOL_RESULT_MIN_STEP_NUMBER.

Verification

  • Headless and desktop context-budget tests for activeToolResultPrune default-on, opt-out, and env knobs.
  • Runtime package typecheck/build and full runtime test suite.
  • Desktop main build/typecheck.
  • Storage package build and AgentRun store tests.
  • Runtime package typecheck/build and focused runtime tests.
  • Storage package build and focused session-store tests.
  • Desktop main build/typecheck and focused bot/OpenGateway, credential-store, settings/web-search, connection IPC, OAuth, and model-provider regression suites.
  • git diff --check before each pushed phase.