Skip to content

README.md

Latest commit

 

History

History
218 lines (155 loc) · 9.17 KB

README.md

File metadata and controls

218 lines (155 loc) · 9.17 KB
Raw

CodeForge

GitHub release License: MIT CI

The open-source AI coding IDE that makes a consumer GPU feel like a coding agent — not a chatbot with tools bolted on.

Built on Code-OSS. Your models. Your machine. No Docker. No vendor lock-in.

Download: Latest Windows installer · Source: extensions/codeforge/

The AHA moment

Most “local AI IDEs” stall when the model is small: they over-read, under-edit, or declare victory without verifying. CodeForge fixes the agent loop itself — read → edit → test → finish — and then adapts the harness to each model.

Measured on our local stack
GPU NVIDIA RTX 3060 12GB
Model Qwen3.6-35B (IQ2_S / ~Q2 GGUF)
Inference SuperLLaMA — usable local llama.cpp on modest GPUs (OpenAI-compatible /v1, TurboQuant KV, Windows installer)
Suite Headless T1–T4 × 3 reps each (benchmarks/)

Harness off (empty learned harness — contract + finish gates only):

Task Mean (3 reps) Stuck badArgs
T1 — bug fix 100 0 0
T2 — feature (coupon) 97 0 0
T3 — multi-file 100 0 0
T4 — rename refactor 100 0 0

Harness on (learned revision 1 for the same model):

Task Mean (3 reps) Stuck badArgs
T1 — bug fix 100 0 0
T2 — feature (coupon) 94 0 0
T3 — multi-file 100 0 0
T4 — rename refactor 100 0 0

Zero stuck runs. Zero bad tool-args. A ~Q2 35B completing real agentic coding tasks end-to-end on a gaming GPU — served by SuperLLaMA, driven by CodeForge's tool contract.

That is the product bet: the model is not the product — the contract around the model is.

Frontier APIs still work. Local weak models stop being a demo and start being a workflow. Pair CodeForge with SuperLLaMA (or any OpenAI-compatible endpoint) and point Settings → AI at your local /v1.

What we actually built

Not another chat sidebar. A tool + system contract that refuses to paper over failures with intent matchers.

Capability Why it matters
Agentic loop that finishes Unlimited exploration within the turn budget (Cursor/Claude Code style). Continue button when the step budget ends. No soft “explore cap” that refuses new reads.
Adaptive harness Per-model profiles (~/.codeforge/ai/model-profiles/) — sampling, write gates, finish limits, optional toolTransport for XML-leaky locals. Calibrate once; hysteresis adjusts online.
Honest outcomes Stuck / truncated / oracle-red are failed, not “done”. Spec self-check and oracle before finish with writes.
You own the freedoms Sandbox level, Allow all, outside-workspace writes, MCP, Full agent freedom — all user settings, revocable. The model cannot grant itself permissions.
Measurable learning Harness revise proposes changes; keep only when paired bootstrap shows a real gain. No quiet contamination of harness.json.
Same IDE for weak and strong Compensations are per model. A clean 32B does not pay the IQ2_S tax.

Where we’re going

Near term

  • Ship releases that treat the adaptive harness + agent loop as the core product story
  • Keep the calibrate suite (T2/T4/CS2/CS3) as the release gate for local models
  • Harden permissions, sandbox, and secrets for daily use without babysitting

Next

  • Broader task suites (navigation, long multi-card, review) with the same failure taxonomy: harness / protocol / semantic
  • Profile knobs proven by A/B (thinking off on tool turns, one mutating call per step) — only when the bench says so
  • Packaging and onboarding so “install → pick local model → agent works” is a five-minute path

Not the roadmap

  • Hardcoded intent crutches (“the user said fix, so nudge edit”)
  • One global “weak mode” that punishes strong models
  • Claiming 100% on every task — residual failures stay visible as model limits, not hidden by prose

Why CodeForge?

CodeForge
Models OpenAI, Anthropic, Gemini, xAI, OpenRouter, Ollama, vLLM, LM Studio, SuperLLaMA / llama.cpp / any OpenAI-compatible API
Adaptive harness Per-model profiles + calibrate + hysteresis
Agent Multi-file loop with approvals; Ask / Plan / Agent + permission chip
Chat Cursor-like sidebar: composer, context ring, tool rows, Undo/Review, Restore, Continue
Sandbox User-configured shell sandbox + optional Full agent freedom
Memory Session wiki, project wiki, stable facts
Governance Skills, rules, policies, hard guardrails
MCP Inbound tools + outbound wiki + Plugins catalog

Also discoverable as: open source Cursor alternative, Ollama coding agent, llama.cpp coding IDE, local LLM agent IDE.

Install (Windows)

  1. Open Releases
  2. Download CodeForge-Setup-*-win32-x64.exe
  3. Run the installer (per-user under %LOCALAPPDATA%\Programs\CodeForge)

Or build from source (below).

Quick start (development)

git clone https://github.com/Kortexio/CodeForge.git
cd CodeForge
npm install
npm run vscode:setup      # clone Code-OSS, brand, install deps
npm run build:extensions # sync calibrate suite + compile AI extension
npm run vscode:compile   # compile editor (first time is slow)
npm run vscode:dev       # launch CodeForge with AI sidebar

Calibrate from the app: CodeForge Settings → Benchmark, or:

node benchmarks/bench.mjs calibrate --model qwen3.6-35b --reps 3
node benchmarks/bench.mjs run T2 --label smoke --model qwen3.6-35b

See docs/code-oss-integration.md and benchmarks/README.md.

Benchmarks (llama.cpp, RTX 3060 12GB)
GPU NVIDIA GeForce RTX 3060 12GB
Server llama.cpp OpenAI-compatible API
Primary local model qwen3.6-35b (Qwen3.6-35B-A3B IQ2_S)
Stronger local Qwen3-32B-exl3-4.0bpw (calibrate A/B)
Runner benchmarks/bench.mjs headless / packaged resources/calibrate-bench/

Agent suite — post harness fix (2026-10-22)

T1–T4 × 3 reps, learned harness off for isolation. Score 0–100.

Model T1 T2 T3 T4 Stuck
qwen3.6-35b (IQ2_S) 100 97 100 100 0

Calibration suite — 2026-10-20

T2 · T4 · CS2 · CS3 × 3 reps × 2 models. Protocol quality decides weak vs strong profile knobs — not the model name. Details: benchmarks/results/CALIBRATE-2026-10-20.md.

Harness design notes: extensions/codeforge/HEURISTICS.md.

Architecture

CodeForge/
├── extensions/codeforge/          # Shipped AI platform (source of truth)
│   ├── src/agent/                 # Loop, profiles, telemetry, revise, tools
│   ├── resources/calibrate-bench/ # Packaged calibrate suite
│   ├── src/views/                 # Chat host + Cursor-like UI
│   ├── src/memory/                # Session + project wiki
│   ├── src/sandbox/               # Shell isolation
│   ├── src/intelligence/          # Index, LSP, git, embeddings
│   ├── src/mcp/                   # Inbound + outbound MCP
│   └── src/governance/            # Skills, rules, guardrails
├── benchmarks/                    # Full harness suite + CLI
├── docs/
└── scripts/                       # Code-OSS setup / package / sync

Configuration

Prefer CodeForge Settings in the app (Models / Permissions / MCP / Guardrails / Agent).

{
  "codeforge.ai.mode": "agent",
  "codeforge.ai.shellSandbox": "restricted",
  "codeforge.ai.allowAll": false,
  "codeforge.ai.outsideWorkspaceWrites": "ask",
  "codeforge.ai.fullAgentFreedom": false
}

Profiles live per machine under ~/.CodeForge/ai/model-profiles/ — calibrate once per model.

Development

Prerequisites: Node matching vscode/.nvmrc, npm, Git. On Windows: VS C++ workload + Spectre-mitigated libs.

npm run build:extensions
npm test
npm run vscode:package:win   # needs Inno Setup

License

MIT — see LICENSE.

Contributing

See Contributing Guide.

Acknowledgments

  • Built on Code-OSS
  • Validated on consumer GPUs (RTX 3060 12GB + llama.cpp) and cloud APIs
  • The agent loop is measured, not marketed — failure classes stay visible