nex-code 0.4.20 → 0.4.22

package/README.md

@@ -22,7 +22,7 @@
   <img src="https://img.shields.io/badge/Ollama_Cloud-supported-brightgreen.svg" alt="Ollama Cloud: supported">
   <img src="https://img.shields.io/badge/node-%3E%3D18-brightgreen.svg" alt="Node >= 18">
   <img src="https://img.shields.io/badge/dependencies-2-green.svg" alt="Dependencies: 2">
-  <img src="https://img.shields.io/badge/tests-3453-blue.svg" alt="Tests: 3453">
+  <img src="https://img.shields.io/badge/tests-3719-blue.svg" alt="Tests: 3719">
   <img src="https://img.shields.io/badge/VS_Code-extension-007ACC.svg" alt="VS Code extension">
 </p>
 
@@ -123,6 +123,10 @@ The verify phase catches incomplete work before reporting "done" — if tests fa
 
 **Lightweight.** 2 runtime dependencies (`axios`, `dotenv`). Starts in ~100ms. No Python, no heavy runtime, no daemon process.
 
+**Server-aware from the first message.** When your prompt contains a URL whose domain matches a configured SSH profile (e.g. `jarvis.example.com` → profile `jarvis`), nex-code probes the server before responding — listing ports, running processes, and data directories. The model receives this topology before its first token, so it goes straight to `ssh_exec` instead of reading local files.
+
+**Few-shot behavior injection.** On each session start, nex-code injects a short example of the correct tool sequence for the detected task type (sysadmin → check remote logs first; coding → read file before editing; data → explain before rewriting). Works across all models without fine-tuning. Customize with your own high-scoring sessions via `npm run extract-examples`.
+
 **Infrastructure tools built in:**
 
 - SSH server management (AlmaLinux, macOS, any Linux)
@@ -143,7 +147,7 @@ The verify phase catches incomplete work before reporting "done" — if tests fa
 
 **Extensible.** Plugin API (`registerTool` + lifecycle hooks), skill system (install from any git URL), MCP server support.
 
-**Tested.** 3453 tests, 79% coverage, CI on every push.
+**Tested.** 3719 tests, 83% coverage, CI on every push.
 
 ---
 
@@ -155,15 +159,33 @@ Rankings are based on nex-code's own `/benchmark` — 15 tool-calling tasks agai
 ### Flat-Rate / Pay-as-you-go
 
 <!-- nex-benchmark-start -->
-<!-- Updated: 2026-03-26 — run `/benchmark --discover` after new Ollama Cloud releases -->
+<!-- Updated: 2026-03-29 — run `/benchmark --discover` after new Ollama Cloud releases -->
 
 | Rank | Model | Score | Avg Latency | Context | Best For |
 |---|---|---|---|---|---|
-| 🥇 | `devstral-2:123b` | **82.5** | 1.7s | 131K | Default — fastest + most reliable tool selection |
-| 🥈 | `devstral-small-2:24b` | 75 | 3.1s | 131K | Fast sub-agents, simple lookups |
-| 🥉 | `qwen3-coder:480b` | 72.5 | 8.4s | 131K | Coding-heavy sessions, heavy sub-agents |
-| — | `kimi-k2:1t` | 67.5 | 6.6s | 256K | Large repos (>100K tokens) |
-| — | `minimax-m2.7:cloud` | 64.1 | 5.0s | 200K | Complex swarm / multi-agent sessions (Toolathon SOTA) |
+| 🥇 | `qwen3-vl:235b` | **77.1** | 14.4s | 131K | Overall #1 — frontier tool selection, data + agentic tasks |
+| 🥈 | `qwen3-vl:235b-instruct` | 76.3 | 6.5s | 131K | Best latency/score balance — recommended default |
+| 🥉 | `rnj-1:8b` | 74 | 3.7s | 131K | — |
+| — | `ministral-3:8b` | 73.1 | 2.3s | 131K | Fastest strong model — 2.2s latency, 70+ score |
+| — | `qwen3-coder-next` | 71.4 | 2.8s | 256K | — |
+| — | `qwen3-next:80b` | 70.6 | 11.6s | 131K | — |
+| — | `qwen3.5:397b` | 68.9 | 3.9s | 256K | — |
+| — | `minimax-m2.7` | 68.7 | 6.8s | 200K | — |
+| — | `glm-5` | 67.6 | 4.5s | 131K | — |
+| — | `devstral-2:123b` | 67.6 | 2.0s | 131K | Sysadmin + SSH tasks, reliable coding |
+| — | `glm-4.7` | 66.5 | 5.1s | 131K | — |
+| — | `kimi-k2-thinking` | 66.3 | 18.4s | 256K | — |
+| — | `ministral-3:14b` | 65.8 | 3.8s | 131K | — |
+| — | `devstral-small-2:24b` | 65.5 | 2.3s | 131K | Fast sub-agents, simple lookups |
+| — | `ministral-3:3b` | 65.4 | 2.2s | 32K | — |
+| — | `kimi-k2.5` | 65.2 | 3.5s | 256K | Large repos — faster than k2:1t |
+| — | `kimi-k2:1t` | 65.2 | 4.2s | 256K | Large repos (>100K tokens) |
+| — | `minimax-m2.1` | 64.2 | 5.4s | 200K | — |
+| — | `glm-4.6` | 63.9 | 4.9s | 131K | — |
+| — | `qwen3-coder:480b` | 63.2 | 14.1s | 131K | Heavy coding sessions, large context |
+| — | `nemotron-3-super` | 61.3 | 2.6s | 256K | — |
+| — | `gpt-oss:20b` | 60.9 | 2.5s | 131K | Fast small model, good overall score |
+| — | `mistral-large-3:675b` | 60.8 | 3.8s | 131K | — |
 
 > Rankings are nex-code-specific: tool name accuracy, argument validity, schema compliance.
 > Toolathon (Minimax SOTA) measures different task types — run `/benchmark --discover` after model releases.
@@ -451,6 +473,8 @@ Fallback chains let you auto-switch when a provider fails:
 /fallback anthropic,openai,local
 ```
 
+**Wire Protocol Layer:** All 5 providers share 3 wire protocol implementations (OpenAI-compatible SSE, Anthropic Messages SSE, Ollama NDJSON). Stream parsing, tool call accumulation, and response normalization are handled by reusable `StreamParser` classes — eliminating duplicated protocol code across providers.
+
 ---
 
 ## Commands
@@ -473,6 +497,12 @@ Type `/` to see inline suggestions as you type. Tab completion is supported for
 | `/load <name>`                                | Load a saved session                                                                                                                               |
 | `/sessions`                                   | List saved sessions                                                                                                                                |
 | `/resume`                                     | Resume last session                                                                                                                                |
+| `/branches`                                   | Show session tree (all conversation branches)                                                                                                      |
+| `/timeline [n]`                               | Show message timeline of current branch                                                                                                            |
+| `/goto <index>`                               | Jump to a message index (truncates later messages)                                                                                                 |
+| `/fork [index] [name]`                        | Create a new branch at the given message index                                                                                                     |
+| `/switch-branch <name>`                       | Switch to a different conversation branch                                                                                                          |
+| `/delete-branch <name>`                       | Delete a conversation branch                                                                                                                       |
 | `/remember <text>`                            | Save a memory (persists across sessions)                                                                                                           |
 | `/forget <key>`                               | Delete a memory                                                                                                                                    |
 | `/memory`                                     | Show all memories                                                                                                                                  |
@@ -860,6 +890,37 @@ Only sessions from the last 24 hours are offered for auto-resume. Older autosave
 
 Sessions are stored in `.nex/sessions/` as JSON files. Auto-saves always write to `_autosave` (overwritten each turn). Writes are atomic — a temp file is written and renamed, so a crash mid-write never corrupts the saved state.
 
+### Session Trees
+
+Navigate your conversation history like git branches. Fork at any point, explore alternative approaches, and switch between branches:
+
+```
+/timeline            # see message indices
+/fork 5 experiment   # branch from message 5
+/branches            # see all branches
+/switch-branch main  # go back to main
+/goto 3              # jump to message 3 (truncates later messages)
+/delete-branch experiment
+```
+
+This enables non-linear conversations: try an approach, and if it doesn't work, fork from an earlier point and try something different — without losing the original attempt.
+
+### Autoresearch
+
+Autonomous optimization loops inspired by Karpathy's autoresearch pattern. The agent edits code, runs experiments, logs results, and automatically keeps improvements or reverts failures:
+
+```
+/autoresearch reduce test runtime while maintaining correctness
+/autoresearch optimize bundle size under 500kb
+```
+
+The agent follows a repeating cycle: **checkpoint** (git) -> **edit** -> **run experiment** -> **log result** -> **keep or revert**. All experiments are logged to `.nex/autoresearch/experiments.json` with metrics and trend tracking.
+
+```
+/ar-status    # show experiment history with trends
+/ar-clear     # reset experiment history
+```
+
 ### Memory
 
 Persistent project memory that survives across sessions:
@@ -1589,10 +1650,12 @@ npm test              # Run all tests with coverage
 npm run test:watch    # Watch mode
 ```
 
-83 test suites, 3453 tests, 79% statement / 71% branch coverage.
+91 test suites, 3719 tests, 83% statement / 74% branch coverage.
 
 CI runs on GitHub Actions (Node 20 LTS).
 
+**Type checking:** `npm run typecheck` runs TypeScript in `noEmit` mode with `allowJs`. Core type definitions live in `types/index.d.ts` (Message, ToolCall, IProvider, IWireProtocol, Session, Skill, etc.). The codebase uses incremental TypeScript adoption — new modules can be written in `.ts` while existing `.js` files are gradually migrated.
+
 ---
 
 ## Dependencies