Releases: sunilp/jam-cli
Releases · sunilp/jam-cli
v0.11.4
v0.11.3
v0.11.2
v0.11.1
v0.11.0 — Windows Support
Windows Compatibility
Jam now works on Windows (PowerShell, cmd.exe, VSCode terminal).
- jam ports — Windows netstat support
- jam vibes — pure Node.js, no Unix shell commands
- jam completion — PowerShell tab completion
- Sandbox — cmd.exe on Windows
- Path display — normalized separators
- npx/npm — .cmd shim resolution on Windows
- Process signals — cross-platform termination
CI
- Windows added to CI matrix
- Homebrew formula auto-updates on release
v0.10.0 — Jam has a personality
Jam has a personality
Jam is no longer a generic AI assistant. It is now the senior dev who has seen everything — direct, opinionated, and warm without being corporate.
Personality across all touchpoints
- JAM_SOUL — central personality definition prepended to all AI system prompts
- System prompts — ask, chat, run, go, planner, worker all speak with jam voice
- Command prompts — trace, verify, commit AI prompts updated
- Error messages — all 25 error hints rewritten to be empathetic and actionable
- Status messages — run and go progress output is direct and concise
- jam doctor — conversational output (Checking your setup... You are good to go.)
- jam init — welcoming first-run (Welcome to Jam. Let me figure out what you have got...)
- Fortune cookies — 8 new developer wisdoms in jam vibes
Fixes
- All ESLint errors resolved (CI was failing on lint)
- tree-sitter parser handles grammar loading failures gracefully
- TypeScript extractor findEnclosingFunction correctly skips variable_declarator for plain assignments
- SQL tests fall back to regex mode when tree-sitter-sql grammar is ABI-incompatible
v0.9.0 — Agent Engine: Orchestrator, Parallel Workers, Workspace Intelligence
Agent Engine
Jam now has a full autonomous agent engine powering both jam go (interactive console) and jam run (one-shot).
Highlights
- Orchestrator — Decomposes tasks into subtasks with a dependency graph, dispatches parallel workers, merges results
- Parallel Workers — Independent subtasks run concurrently (up to 3 by default, configurable with
--workers) - Workspace Intelligence — Profiles your project code style, test patterns, naming conventions, and frameworks before writing code. Cached in
.jam/workspace-profile.json - Tiered Permissions — Commands classified as safe/moderate/dangerous with automatic approval or confirmation
- OS Sandbox —
sandbox-exec(macOS) /unshare/firejail(Linux) restricts filesystem access to workspace - Multimodal Input —
--imageflag for vision-capable models (OpenAI, Anthropic, Gemini, Ollama/llava) - File-Lock Protocol — Prevents parallel workers from conflicting on shared files, with deadlock detection
New Commands & Flags
jam go— Interactive agent console with/stop,/status,/help,/exitjam go --auto— Fully autonomous modejam run --auto— Autonomous one-shot (implies--yes)--workers <n>— Max parallel workers--image <path>— Attach image for vision models--no-sandbox— Disable OS-level sandbox--file <path>— Read task from file (jam run)JAM_LEGACY_RUN=1— Fallback to pre-0.9 run behavior
Stats
- 28 new files in
src/agent/ - 158 new tests (811 total)
- Zero new dependencies
v0.8.7 — jam go: Interactive Agent Mode
New Features
jam go— Claude Code-like interactive agent session- Full write tools: write_file, apply_patch, run_command, git operations
- Reads your codebase, plans changes, writes code, runs tests
- Permission enforcement via toolPolicy config
- Streaming output with tool execution indicators
- Session persistence
- Ctrl+C to interrupt, double Ctrl+C to exit
Bug Fixes
- verify, diff, review, commit, patch now work with Copilot proxy (removed incorrect tool-support guard)
- Only
jam runrequires tool-calling provider support
v0.8.6 — Fix: verify, diff, review, commit, patch work with Copilot proxy
Bug Fixes
- Removed incorrect tool-support guard from verify, diff, review, commit, and patch commands
- These commands only use streaming (not tool calling) and now work correctly with the Copilot proxy backend
- Only
jam runretains the tool-support guard since it actually uses tool calling
v0.8.5 — VSCode Context: explain this, selection, git root
New Features
jam explainwith no args uses the active VSCode editor filejam explain thisworks naturally in VSCode terminal- Selected text in VSCode editor is explained when using
jam explain - All git commands (
jam review,jam commit,jam diff, etc.) now resolve the correct git root from the active editor file — fixes 'Failed to determine current branch' errors in multi-repo workspaces - New
/v1/contextendpoint on the VSCode proxy server
Bug Fixes
jam doctorshows keytar/ripgrep as warnings [!] instead of failures [x]- Copilot adapter lazily initializes — all commands work without explicit validateCredentials()