Maka

A local-first Agent workspace built for real work.

Maka does more than answer questions. With controlled permissions, it can inspect projects, execute tools, produce artifacts, and preserve model messages, tool calls, and durable-task progress as recoverable execution facts. The same Runtime is available through the desktop app, terminal TUI, non-interactive CLI, and Headless runner.

Important

Maka is under active development. The macOS Apple Silicon desktop build is an early public release; data formats, CLI commands, and experimental capabilities may still change.

Why Maka

• Local-first instead of hosted-first: sessions, settings, and run records stay on your machine by default. You choose the model connection: cloud API, local model, or compatible gateway.
• Log is the Runtime: model messages, Tool Calls, Tool Results, and termination facts enter Runtime Event Log. Sessions, UI, model context, and recovery are projections over that log.
• Context is not history: Tool Result pruning and LLM Compaction change what the next inference sees without treating recorded evidence as disposable context.
• A task may outlive a Turn: Headless uses TaskRun, Task Event Log, budgets, and continuation to advance interruptible and inspectable durable work.
• Feedback is not fact authority: Self-check may produce evidence and one bounded repair opportunity, but “I checked it” does not become a system fact.

Read Maka Backend Architecture for the complete design.

Surfaces

Entry point	Best for	Current capability
Desktop	Daily interaction, file and Artifact workflows, model and permission setup	Electron + React with streaming sessions, tool timelines, branching, search, and recovery
TUI / CLI	Using Maka in the current project directory or running one non-interactive Turn	maka, maka run; shares workspace and model connections with Desktop
Headless	Durable tasks, recoverable TaskRuns, experiments, and evaluation	maka eval with task logs, export, resume, and comparison

Current capabilities

Agent Runtime

• Multiple model connections, streaming output, thinking, usage accounting, and provider-error normalization;
• Local tools including Read, Write, Edit, Bash, Glob, and Grep;
• Tool schema validation, dynamic availability, permission policy, watchdogs, abort, and error classification;
• Runtime Event Log, AgentRun ledger, startup recovery, Turn Evidence, active Tool Result pruning, and history compaction.

Desktop workspace

• Create, archive, search, rename, retry, regenerate, and branch sessions from a Turn;
• Artifact lists and previews, workspace instructions, model settings, and permission settings;
• Local memory, web search, and bot entry points;
• Integrations are configured independently, and not every experimental entry is available by default.

Durable tasks and evolution

• Append-only Task Event Log and TaskRun projection;
• Budgets, permission pauses, continuation, result export, and failed-task retry;
• Plan-first, source-guarded, and attempt-bounded Heavy-task Self-check;
• AHE target protocol and evidence export; complete automatic self-iteration remains an external or experimental workflow.

Quick start

Download Desktop for macOS

The signed and notarized Desktop app is available from GitHub Releases for Apple Silicon Macs only (arm64).

1. Download Maka-<version>-mac-arm64.dmg;
2. Open the DMG and drag Maka to Applications;
3. Install ripgrep with brew install ripgrep to enable Runtime's Grep tool;
4. Launch Maka and configure your own model connection under Settings → Models.

Computer Use is not included in this first public build. Intel Macs, Windows, and Linux packages are not supported yet.

Requirements

• Node.js 22.19 or newer (CI uses Node.js 24);
• npm (the lockfile and scripts use npm; the current packageManager is npm 11);
• Git;
• ripgrep, used by Runtime's Grep tool.

Start Desktop

git clone https://github.com/Maka-Agent/maka-agent.git
cd maka-agent
npm ci
npm run dev

npm run dev starts the Desktop development environment with HMR. To build every workspace before starting Electron, use:

npm run dev:full

If dependencies were installed with ELECTRON_SKIP_BINARY_DOWNLOAD=1, install the Electron platform binary before starting:

node node_modules/electron/install.js

First run

Maka does not bundle a shared model account. On first launch:

1. Open Settings → Models;
2. Add an API, local-model, or supported account connection;
3. Test it and choose a default model;
4. Return to the workspace and start a task.

The app distinguishes configured, send-ready, and experimental connection states. An account flow that is not wired into Runtime is not presented as a usable model.

Terminal entry points

Build the workspaces first:

npm run build

Then start the TUI or run one Turn:

npm --workspace maka-agent exec -- maka
npm --workspace maka-agent exec -- maka run "Summarize this repository and identify its most important risk"
npm --workspace maka-agent exec -- maka run --graph "Implement two independent slices, integrate them, then review the result"
npm --workspace maka-agent exec -- maka --help

The TUI also accepts /graph on, /graph off, and /graph <task>. Non-interactive --graph runs wait for the durable Graph to finish before printing the final supervisor output. Graph implementation operators use isolated Git worktrees, so the source project must be a clean Git worktree.

The CLI reads the same model connections and workspace configuration written by Desktop. See packages/headless/README.md for Headless commands and its trust posture.

Architecture

The backend spine is:

Desktop / TUI / Headless
          ↓
SessionManager → AgentRun → Model + Tool Runtime
          ↓
Runtime Event Log → Context / Session / UI projections
          ↓
Task Event Log → TaskRun → Self-check / AHE evidence

Start with ARCHITECTURE.md. It provides the system map, code boundaries, problem-oriented reading paths, and six bilingual deep dives.

Repository layout

apps/desktop/       Electron main / preload / React renderer

packages/core/      Pure contracts for Sessions, Events, Permissions, and Connections
packages/storage/   SQLite operational state, legacy importers, and payload stores
packages/runtime/   AgentRun, model adapters, tools, context, and recovery
packages/headless/  TaskRun, Autonomous Loop, Self-check, eval, and AHE
packages/cli/       TUI and non-interactive CLI
packages/ui/        Shared conversation, Markdown, Artifact, and UI primitives

docs/               Architecture, product, security, privacy, and test contracts
scripts/            Build hygiene, visual checks, smoke tests, and release helpers

Local data and security boundary

Maka stores workspace data under Electron userData by default:

<Electron userData>/workspaces/default/
  llm-connections.json
  credentials.json
  settings.json
  sessions/

Current boundaries that matter:

• Sessions and connection metadata live in the local filesystem;
• Runtime credentials such as API keys, bot tokens, and proxy passwords currently live in local plaintext credentials.json, behind the OS account boundary, with POSIX directory mode 0700 and file mode 0600 enforced;
• Subscription OAuth tokens (Claude, Codex, GitHub Copilot, and the Cursor/Antigravity previews) live in the same credentials.json — the single authority for desktop, TUI, and headless; Electron safeStorage only decrypts pre-existing legacy token files once at desktop startup (#1125);
• Renderer does not receive plaintext credentials. File writes, Shell, and dangerous tool calls pass through the permission engine;
• Headless real-model evaluation fails closed by default and requires an explicit external isolation boundary.

Read SECURITY.md for security reporting and policy, and docs/README.md for current privacy and sandbox contracts.

Runtime storage and recovery

RuntimeEvent persistence is always canonical in runtime.sqlite. On the first write, Maka batch-idempotently imports legacy RuntimeEvent JSONL without rewriting it. Legacy-only workspaces remain available to read-only inspection until that first write.

Session metadata and Agent Graph control tables now use the same process-local operational database owner and the same runtime.sqlite transaction authority. The first operational open copies a WAL-consistent sessions.sqlite source into runtime.sqlite, validates every source row, and records the source digest and result in cutover_journal. An interrupted copy resumes without partial rows; a legacy database changed after cutover fails closed. The old database is retained as migration evidence but is no longer a production writer. Session transcript bodies remain append-only JSONL.

Core execution state now shares that authority too: AgentRun headers and event ledgers, event projections, root-turn admissions and source proofs, Interactions, Host Epoch message receipts, and ShellRun records are canonical in runtime.sqlite across CLI, Desktop, Runtime Host, and Headless. Each legacy file store is fingerprinted and imported through its own durable cutover_journal entry before the corresponding repository opens. Copy and validation are one SQLite transaction, retries are idempotent, and a changed legacy source after cutover fails closed. The legacy files are retained only as migration evidence; new execution writes do not modify them.

Workflow state is migrating in reviewable slices. Task Ledger events and projections, Plan events and projections, Deep Research events, and Plan Reminder records are now canonical in runtime.sqlite. Desktop and Runtime Host production wiring opens these SQLite repositories; their JSON/JSONL predecessors are read only during a fingerprinted, crash-safe cutover and are never updated by later mutations.

Usage telemetry and pricing authority now use that same operational database. Legacy telemetry.json and pricing.json sources are decoded together and fingerprinted before their rows and pricing revision are committed atomically. After cutover, Desktop and Runtime Host write only runtime.sqlite; the source files remain unchanged as migration evidence.

Artifact metadata and lifecycle state now follow the same rule. Payload bytes remain files, but their records are canonical in runtime.sqlite after a fingerprinted metadata.jsonl cutover. Payload publication keeps its durable staging/link protocol: recovery removes bytes whose metadata transaction did not commit and preserves committed bytes while cleaning staging residue. Purge intent recovery likewise completes against the SQLite metadata authority.

Selected-session bundle export now reads that SQLite authority through a WAL-consistent snapshot, verifies that retained legacy evidence still matches its completed cutover, and writes only the selected Artifact rows into the bundle's runtime.sqlite. Payload bytes are copied under the Artifact writer lock, cross-session rows and payloads are excluded, and bundles no longer emit artifacts/metadata.jsonl.

Full operational backup now uses the shared database owner's online SQLite backup API rather than copying runtime.sqlite or its WAL sidecars. A strict manifest binds the standalone database snapshot to every active session transcript and canonical Artifact payload by size and SHA-256. Restore verifies SQLite integrity, foreign keys, supported schema versions, relational identity sets, transcript decodability, and the exact manifested file tree before atomically publishing a new state root. Interrupted backup or restore staging is removed and can be retried without changing either source.

Headless trajectory hydration now consumes a frozen selected-session export from that SQLite Artifact authority. The cell publishes trajectory-state only when RuntimeEvents reference image Artifacts; Harbor downloads its standalone runtime.sqlite first and then only the payloads referenced by the validated snapshot. It does not copy a live WAL or fall back to artifacts/metadata.jsonl. Missing, corrupt, unsupported, or mismatched evidence fails closed to a summary trajectory instead of mixing authorities.

The remaining storage work is deliberately classified rather than implied complete:

• Artifact metadata no longer exposes a production JSONL writer; artifacts/metadata.jsonl is accepted only as fingerprinted, read-only cutover evidence, and can be removed after the migration/cutover matrix is complete;
• StoredMessage transcript bodies remain append-only JSONL;
• automation, connections, credentials, settings, MCP configuration, skills, and device identity are configuration state and stay outside this operational migration;
• Headless TaskRun evaluation ledgers, imported foreign-session caches, Daily Review archives, and quote-cleanup bookkeeping are separate product/evaluation domains and are not part of issue #1649.

Runtime continuation remains opt-in:

• MAKA_RUNTIME_SAFE_BOUNDARY_RESUME=1 enables the Desktop interrupted-turn Safe resume action, CLI/TUI /resume, and Desktop startup auto-resume. These paths may call the configured model provider and consume tokens. Enable the flag only when that behavior is explicitly desired.

Phase 2 provides the durable write-side boundary and fail-closed safe-boundary continuation. Phase 3 reconciliation for indeterminate tool side effects is not implemented yet; ambiguous tool outcomes remain parked rather than retried.

Development and verification

Common repository-level commands:

npm run build
npm run typecheck
npm test
npm run check:release

Run one workspace in isolation:

npm --workspace @maka/runtime test
npm --workspace @maka/headless test
npm --workspace @maka/desktop test

Use the following commands to update packages/core/src/model-metadata.generated.ts from models.dev and run the focused tests. Keep access-path-specific overrides in model-metadata.ts; do not edit the generated file by hand.

npm run sync:model-metadata
npm run test:scripts
npm --workspace @maka/core test

Desktop real-window and visual verification:

npm --workspace @maka/desktop run e2e
npm --workspace @maka/desktop run smoke:real-window

Before submitting code, run typecheck, build, and focused tests proportionate to the change, followed by git diff --check.

Documentation

• Documentation index and authority map
• Backend architecture
• Product design
• Security policy

License

Maka is licensed under the Apache License 2.0. See NOTICE for attribution information. Third-party components remain subject to their respective licenses and notices.