Ogcode

The token-efficient agentic coding workbench.

Built for a future where every token counts. Ogcode curates the relevant context for each turn — not the full transcript — so it cuts 70%+ of tokens on long sessions, sharpens accuracy, and lets even lower-end models outperform frontier ones. And because it recalls instead of replays, your conversations run effectively forever — you never hit a model's context limit, on any model, frontier or local. All while planning with you, remembering your codebase, and shipping features in parallel from a single binary that never leaves your machine.

Ogcode Demo

Context Engineering · Infinite Context · Plan Mode & Parallel PRs · Quick Start · Why Ogcode · Documentation · Discord

Context Engineering — the real differentiator

> Every other coding agent resends your entire conversation history on every turn. Ogcode doesn't.

Most coding agents operate on a naive replay loop: each turn, they bundle up the full transcript so far — every prior message, every tool result, every tangent — and ship it back to the model. That has two costs, and only one of them is money.

1. It burns tokens. The prompt grows linearly with the session, so a 200-message task can cost 5× more than a 20-message one even if the new work is trivial. On a fixed monthly budget this caps how much you can actually ship.

2. It hurts accuracy — and this matters more than the money. An LLM can only act on what's in its context window. When you flood that window with stale, unrelated chatter from earlier in the session, the signal gets buried in noise: the model loses sight of the current task, drifts toward half-remembered earlier decisions, and reasons against context that was relevant then but isn't relevant now. The older the conversation, the more the historical turns actively distract from the turn in front of the model.

Ogcode does the opposite. For each turn it extracts only the context that is actually relevant to the task at hand — pulling precise facts from a persistent knowledge graph and call graph via memory_recall, fetching code-structure context on demand, and compacting stale history instead of replaying it verbatim. The model receives a short, sharp, on-point context window. Less history, fewer tokens — and better outcomes, because the model isn't wading through a hundred old messages to find the three facts it needs right now.

Saving tokens isn't only about cost — it's about accuracy. A smaller, more relevant context window lets the model focus, so it produces more correct, more on-target results per turn. The two goals reinforce each other.

This is unique to Ogcode

No other coding agent on the market does this. Claude Code, Cursor, Copilot, Aider — every one of them replays the full conversation every turn. Token efficiency, in those tools, is an afterthought at best. Ogcode is the only agent engineered, at the agent-loop level, to conserve tokens and to curate context per turn — because it believes the real lever is context engineering: how efficiently and how relevantly you prepare the context for a given task.

Context engineering lets smaller models punch above their weight

This is the deeper payoff. If the only thing context engineering did was save money, it would still be worth it — but it does more: it extracts capability even from lower-end models. Keep the context relevant, limited, and short, and a mid-tier model (Claude Sonnet, a local Llama, a smaller GPT) can reason just as clearly — and sometimes outperform — a frontier model that's been handed a bloated, noisy transcript. The frontier model isn't smarter about your code; it just has more raw capacity to dig itself out of the irrelevant history you buried it under. Give either model a clean, on-point context window and the gap narrows dramatically — often to zero.

In the end, it's all about context engineering. Ogcode is brilliant at this, which is why it simultaneously cuts token cost and increases the accuracy of the task outcome. Cheaper and better — not a tradeoff.

Infinite Context — never hit a model's limit

> Every model has a context window. Every other agent eventually slams into it. With Ogcode you never do — chat forever, on any model, no matter how small its window.

This is the part that genuinely changes the game. Every LLM ships with a fixed context limit — 8K, 128K, 200K, a million tokens — and every other coding agent creeps toward that wall as the session grows, until the model starts dropping the start of the conversation or simply refuses to continue. Ogcode removes the ceiling entirely.

The reason is Agentic Session Memory. Because Ogcode recalls the few facts relevant to the current turn from a persistent knowledge graph — instead of replaying the entire transcript — the prompt it sends stays flat no matter how long the conversation runs. A session that's 50 messages deep and one that's 5,000 messages deep hand the model the same compact, on-point context window. The conversation is unbounded; the per-turn context is not. So you can talk to a model forever and never reach its limit — and that holds for any model, whatever the size of its native window.

Three wins from one idea

Infinite conversations, on any model. Drive Ogcode with a frontier model or a tiny local Llama with an 8K window — it makes no difference. The knowledge graph keeps every turn small, so the model's own context limit is never the bottleneck. You get effectively limitless session length even on models that could never sustain it alone.
Far lower cost. A flat per-turn prompt means you stop paying the linearly-growing bill of full-transcript replay — over 70% fewer tokens on long sessions. See Context Engineering and Token Efficiency.
Higher accuracy. The model reasons over only the facts that matter for the turn in front of it, not a wall of stale history — so it drifts less and stays on-target. Cleaner context is more accurate, not just cheaper.

Never hit a context limit, spend far fewer tokens, and get more accurate results — on whatever model you choose. That is the true beauty of Agentic Session Memory. (For the knowledge-graph internals, see Agentic Session Memory.)

Plan Mode + Parallel PRs — ship features, not just suggestions

> Other agents suggest code. Ogcode plans the feature, decomposes it, executes the pieces in parallel, and raises the pull requests for you.

Most coding agents stop at "here's the code for the file you asked about." Ogcode's Plan Mode turns a one-line goal into a shipping feature. You describe what you want to build; the planning agent reads your codebase, discusses the approach with you, and — once you lock the plan — it becomes Ogcode's responsibility to break that feature into smaller, implementation-ready tasks, run them, and open the pull requests.

Plan once, ship the whole feature

┌──────────────┐    ┌──────────────┐    ┌──────────────┐    ┌──────────────┐
│  1. Describe │ → │  2. Lock      │ → │  3. Review    │ → │  4. Execute   │
│   your goal   │    │   the plan    │    │  Kanban board │    │   in parallel │
└──────────────┘    └──────────────┘    └──────────────┘    └──────┬───────┘
                                                                    │
                    ┌──────────────┐    ┌──────────────┐    ┌───────▼──────┐
                    │  6. Retry    │ ← │  5. Complete  │ ← │  Task runs    │
                    │  if needed    │    │  auto-PR      │    │  in isolated  │
                    └──────────────┘    └──────────────┘    │  git branch   │
                                                            └──────────────┘

Describe — Open a plan and describe your goal. The planning agent reads your codebase and refines the approach with you in conversation.
Lock — When you're happy, lock the plan. Ogcode runs a breakdown agent that turns the agreed plan into a structured task list with effort estimates (S/M/L/XL), complexity scores, and a dependency graph.
Review — Tasks land on a visual Kanban board. Inspect, reorder, or kick off tasks individually.
Execute — Each task gets its own git branch and isolated agent session. Tasks with no dependency on each other run in parallel — multiple agents coding at the same time.
Complete — Finished tasks auto-commit, push, and open pull requests directly against your GitHub repo.
Retry — A failed task is retried from a clean slate: the stale branch is deleted and the task starts fresh.

How Ogcode breaks the feature down

The breakdown is deliberate about parallelism and conflict-free merges — this is the part most agents get wrong and Ogcode gets right:

File-ownership rule. The breakdown agent is instructed that parallel tasks must not touch the same files. If two tasks would edit the same file, the agent adds a dependency between them so they run sequentially instead. This is what keeps the auto-PRs merge-conflict-free.
Strictly linear dependency chains. Each task may depend on at most one other task, producing clean A→B→C chains rather than tangled fan-in graphs. Cycles are detected and rejected before execution ever starts.
Implementation-ready descriptions. Every task description names exact file paths, function/type names, patterns to follow, and edge cases — so a developer (or the task agent) can implement it from the description alone.
Shared chain branches. Tasks in a dependency chain share one branch and accumulate each completed step, so a downstream task always sees its predecessors' work. The chain raises a single consolidated PR when the last task completes, instead of N conflicting ones.

Why this matters

The payoff is end-to-end parallel feature completion: you plan one feature, and Ogcode ships it as a set of clean, reviewable pull requests that don't fight each other. It does this because git is baked into the Ogcode core, not bolted on as an afterthought:

Git worktrees at the core. Every task runs in an isolated git worktree — a real, independent working checkout on its own branch. Multiple agents work simultaneously without clobbering each other's files or your main branch.
Auto-commit, push, and PR. When a task finishes, Ogcode commits any leftover changes, pushes the branch to origin, and opens a pull request via the gh CLI — idempotently (re-running won't create duplicate PRs), and with a generated PR body.
Stacked PRs, handled gracefully. Dependent tasks branch from their predecessor's branch, not from HEAD. If a stacked base branch was already merged and deleted, Ogcode falls back to the repo's default branch automatically so the PR is still created correctly.
Merge-conflict-safe by construction. Because parallel tasks own disjoint files (enforced at breakdown time) and chained tasks share a branch they merge into sequentially, the PRs Ogcode raises are designed to merge without conflicts.
Beautiful cleanup. Worktrees, branches, and stale directories are pruned automatically — on success, on failure, on retry, and on plan deletion. Crashed sessions are recovered and cleaned up on the next server start.

In short: you describe the feature, Ogcode plans it, splits it into parallel-safe tasks, executes them across isolated branches, and opens the PRs — directly to your upstream GitHub repo, with no merge conflicts and no manual git wrangling. That's parallel task completion and parallel feature completion, from a single binary that manages the GitHub repo and its git worktrees natively because git is part of the Ogcode core.

Ogcode is an agentic coding assistant that runs entirely on your machine — a single Go binary with an embedded SolidJS web UI. It doesn't just suggest code. It understands your whole codebase, plans complex features with you, and executes them across parallel git branches — while your code stays local and private.

Unlike IDE-locked assistants (Cursor, Copilot) or cloud-only services (Claude Code), Ogcode is browser-native, self-hosted, and model-agnostic. Use Claude, GPT, OpenRouter, or local Ollama models, and switch anytime from the UI. No subscriptions. No vendor lock-in. Nothing leaves your machine except the prompts you send to your chosen provider.

curl -fsSL http://ogcode.xyz/install.sh | sh &amp;&amp; ogcode

How Ogcode Compares

	Ogcode	Cursor	Claude Code	Copilot	Aider
Interface	Web UI (any editor)	VS Code fork	Terminal	IDE extension	Terminal
Self-Hosted	Single binary, zero deps	Cloud-required	Cloud-only	Cloud-only	Open source
Parallel Tasks	Git worktrees, conflict-free auto-PRs to upstream	Cloud agents	Subagents	Single agent	Sequential
Plan Mode	Kanban + effort estimates	Agents window	Architect mode	Prompt-based	`/architect`
Persistent Memory	Knowledge graph + Call graph	Session-only	CLAUDE.md	None	None
Token Efficiency	Loop-level optimization, ~70% saved, higher accuracy	No	No	No	No
Model Choice	Claude, GPT, OpenRouter, Ollama	Built-in + custom	Claude only	MS-managed	Any endpoint
Cost	BYOK (tokens only)	$20–$40/mo	$20–$100/mo	$19–$39/mo	Free (BYOK)
License	MIT	Proprietary	Proprietary	Proprietary	Apache-2.0

Ogcode is the only agentic coding assistant that combines a browser-native UI (works with Vim, Emacs, VS Code, JetBrains, or any editor), a formal Plan Mode with a visual Kanban board and parallel execution that raises conflict-free PRs directly against your upstream GitHub repo, git-native parallel execution that gives every task its own isolated worktree branch with auto-commits and auto-PRs, a persistent knowledge graph for long-term memory, loop-level token optimization that keeps long-session token use over 70% lower than naive replay and sharpens per-turn accuracy, context engineering that lets lower-end models match or beat frontier models on clean context, and single-binary self-hosting with zero cloud dependencies.

Features

Build Mode + Plan Mode — Chat with a coding agent in real time, or collaboratively plan complex features with effort estimates and dependency graphs, then let Ogcode decompose them into parallel tasks and raise the PRs.
Token-Efficient by Design — Token optimization is built into the agent loop: the knowledge graph recalls only relevant context per turn, call-graph facts are fetched on demand, and stale history is compacted instead of re-sent — saving 70%+ of tokens on long sessions and sharpening accuracy by keeping the context window focused on the task at hand.
Parallel Task Execution — Independent tasks run simultaneously across isolated git worktree branches and open pull requests directly against your upstream GitHub repo. Ship entire features in parallel, with conflict-free PRs by construction.
Agentic Session Memory — Infinite context via a persistent knowledge graph, with ~70% token savings on long sessions.
Knowledge Graph + Call Graph — Semantic memory of your codebase (Topic → Concept → Fact) plus function-level call relationships for intelligent navigation.
Multi-Provider LLM Support — Anthropic Claude, OpenAI GPT, OpenRouter, or local Ollama models. Switch anytime from the UI.
Deep Research Agent — A built-in deep_search tool searches the web, fetches pages, and synthesizes cited research for your agent.
Kanban Board — Visual task board with S/M/L/XL effort estimates, complexity scores, and dependency chains.
Permission-Based Safety — Destructive operations (write, edit, bash) require explicit approval per tool; read-only tools auto-approve.
PDF Support — Read and index PDF documentation directly in the agent.
Codebase Map — A semantic index of your entire project for intelligent file discovery.
Single Binary, Self-Contained — One Go binary with an embedded SolidJS frontend. Zero external server dependencies.
MCP Extensibility — Model Context Protocol support for custom tools and integrations.
Rich Visual Output — Agents render Mermaid diagrams, LaTeX math, Plotly charts, and full HTML/CSS/JS interactive content directly in the chat, with viewport-aware responsive design.

Quick Start

# macOS / Linux — one-line install
curl -fsSL http://ogcode.xyz/install.sh | sh

# Set your API key (or use Ollama for local models)
export ANTHROPIC_API_KEY=sk-ant-...

# Start coding
ogcode

Opens at http://localhost:9595. That's it — no config files, no Docker, no IDE extension.

Why Ogcode

Cursor is great, but it's a fork of VS Code. If you use Vim, Emacs, or JetBrains, you're out of luck — and its parallel agents run in the cloud, not on your machine.

Claude Code is powerful, but it's terminal-only, Anthropic-only, and cloud-only. No web UI, no plan mode, no persistent knowledge graph.

GitHub Copilot is everywhere, but it's a Microsoft service. Your code analysis happens in the cloud, with no parallel execution and no formal planning.

Aider is excellent, but it's terminal-only and sequential, with no persistent memory graph or visual planning board.

Ogcode gives you what none of the above do:

Context engineering — only the relevant context is sent per turn, so tokens fall and accuracy rises; smaller models can match or beat frontier models on clean context
A web UI that works with any editor
Formal Plan Mode with a visual Kanban board, parallel task decomposition, and auto-PRs raised directly against your upstream GitHub repo
A persistent knowledge graph that survives across sessions
Single-binary self-hosting with zero cloud dependencies
Full model freedom — Claude today, GPT tomorrow, local Llama next week

Token Efficiency

> Nobody else is thinking about token optimization. They burn tokens like water — Claude Code, OpenCode, GitHub Copilot, every coding agent out there — because none of them are designed, at the core loop level, to actually conserve tokens. And because they resend the whole transcript every turn, they also hand the model a noisier, less accurate context window.

As the cost of using frontier AI climbs with every intelligence leap, tokens are becoming a budgeted resource. In the near future, a team — or a solo developer — will have a fixed monthly token allowance and have to ship software, features, and fixes within it. Ogcode is built for that future: token efficiency is designed into the agent loop itself, not bolted on after the fact — and it doubles as an accuracy win, because the model reasons over a curated, on-point context window instead of a wall of stale chat.

How Ogcode saves tokens

Mechanism	What it does	Token impact
Agentic Session Memory	Replaces "send the whole conversation every turn" with a knowledge graph that returns only the facts relevant to the current query.	Largest single saving — grows with session length
Call Graph recall	Pulls in code-structure context on demand instead of re-reading source files into the prompt.	Avoids re-sending large file contents
Context compaction	Summarizes stale history instead of replaying it verbatim, with truncation as a fallback.	Caps prompt size on long sessions
Targeted `memory_recall`	The agent retrieves precise historical facts (config values, past decisions) rather than re-deriving them by re-reading code.	Fewer exploration turns

Real session results

In real session testing, Ogcode saves over 70% of tokens on long-running sessions versus a naive full-replay loop — meaning a fixed monthly budget goes further, a team stays under its limit, and frontier-model cost increases hurt less. And because the model sees only the relevant facts for the current turn rather than the entire transcript, task accuracy improves at the same time: less drift, fewer half-remembered earlier decisions, more on-target results.

Session Length	Traditional (full replay)	With Ogcode memory	Savings
50 messages	~25K tokens	~8K tokens	68%
200 messages	~100K tokens	~28K tokens	72%
1000 messages	~500K tokens	~120K tokens	76%

Enable it with:

export OGCODE_AGENTIC_MEMORY_MODE=true

See Agentic Session Memory for the technical deep dive.

System Requirements

Platform	Minimum
Operating System	macOS, Linux, or Windows
Go	1.22+ (for `go install` only)
Git	2.34+ (required for worktree support)
CPU	Any modern x86_64 or arm64 processor
Memory	512 MB free RAM

An LLM API key or a local Ollama installation is required. See Configuration.

Installation

macOS / Linux

Via Homebrew (recommended):

brew tap prasenjeet-symon/ogcode
brew install ogcode

Via curl (one-liner):

curl -fsSL http://ogcode.xyz/install.sh | sh

Auto-detects your platform, downloads the latest release, and installs to /usr/local/bin.

Windows

irm http://ogcode.xyz/install.ps1 | iex

Downloads the latest release, extracts to %LOCALAPPDATA%\ogcode, and adds it to your PATH.

Via winget:

winget install prasenjeet-symon.ogcode

Go Install

go install github.com/prasenjeet-symon/ogcode@latest

Docker

docker run -p 9595:9595 -v $(pwd):/workspace -w /workspace ghcr.io/prasenjeet-symon/ogcode:latest

Configuration

Ogcode auto-detects available AI providers from environment variables. No config files required.

AI Provider

Set at least one API key (or use Ollama):

Variable	Provider
`ANTHROPIC_API_KEY`	Anthropic (Claude)
`OPENAI_API_KEY`	OpenAI (GPT)
`OPENROUTER_API_KEY`	OpenRouter
`OLLAMA_BASE_URL`	Ollama (local / cloud URL)

Ollama (local models — free, private)

# macOS / Linux — auto-detected if ollama is installed
ollama serve
ogcode

# Or explicit on any OS:
export OLLAMA_BASE_URL=http://localhost:11434/v1
ogcode

Available models: qwen3, codellama, llama3.1, deepseek-coder-v2, mistral, and any model you've pulled.

Agentic Memory (optional)

Enable infinite-context memory across sessions:

export OGCODE_AGENTIC_MEMORY_MODE=true

Web Search (optional)

Give your agent the ability to research current documentation:

export OGCODE_SEARCH_ENABLED=true

Usage

Build Mode (default)

ogcode

Chat with the agent — ask it to read files, write code, run commands, or search the codebase.

Plan Mode

ogcode plan

Describe what you want to build. The planning agent reads your codebase, discusses the approach, and breaks it into tasks with dependencies and effort estimates. Lock the plan, and the tasks become a Kanban board you can execute in parallel. Completed plans are archived as markdown in .ogcode/archives/.

Custom port

ogcode -p 3000
ogcode plan -p 3000

Remote Deployment

Ogcode is just an HTTP server — host it on a remote machine and reach it from any browser.

Expose the port

docker run -p 9595:9595 \
  -v ~/.ogcode:/root/.ogcode \
  -v $(pwd):/workspace -w /workspace \
  ghcr.io/prasenjeet-symon/ogcode:latest

Then open http://:9595 from any browser.

Behind a reverse proxy with HTTPS

# nginx config
server {
    listen 443 ssl;
    server_name ogcode.yourdomain.com;

    location / {
        proxy_pass http://127.0.0.1:9595;
        proxy_set_header Upgrade $http_upgrade;     # WebSocket support
        proxy_set_header Connection "upgrade";
    }
}

Access via https://ogcode.yourdomain.com — encrypted and clean.

Docker Compose

services:
  ogcode:
    image: ghcr.io/prasenjeet-symon/ogcode:latest
    volumes:
      - ~/.ogcode:/root/.ogcode
    ports:
      - "127.0.0.1:9595:9595"   # only localhost — nginx handles public access
    restart: unless-stopped

Security considerations

Ogcode is a coding agent that can execute shell commands, read and write files, and modify your system. Never expose it to the public internet without authentication.

Risk	Mitigation
Anyone can hit port 9595	Bind to `127.0.0.1` + use a reverse proxy
No auth on the web UI	Add HTTP Basic Auth in nginx, or use a VPN
Full shell access via the agent	Run in a restricted environment (Docker, VM)

Recommended approaches:

SSH tunnel — most secure, zero config:

# On your laptop:
ssh -L 9595:localhost:9595 user@your-server
# Then open http://localhost:9595 in your browser

nginx + HTTP Basic Auth — a simple password gate:
```
htpasswd -c /etc/nginx/.htpasswd your_username
```
Cloudflare Tunnel — zero open ports; add Cloudflare Access for auth.
VPN (WireGuard / Tailscale) — a private network with no public exposure.

Rich output

Agents can produce rich visual content directly in the chat — not just plain text:

Format	Syntax	Use For
Mermaid diagrams	```mermaid	Flows, architectures, sequences, ER diagrams
LaTeX math	$...$ or `$$...$$`	Mathematical formulas and equations
Plotly charts	```plotly	Bar, line, scatter, pie, heatmap, and more
Rough diagrams	```rough	Hand-drawn style 2D diagrams
HTML/CSS/JS	```html	Interactive dashboards, styled tables, animations

HTML blocks render in a sandboxed iframe — scripts run in isolation with no access to the parent page. The agent is given your viewport dimensions so it can design responsive content that fits your screen.

Agentic Session Memory

Agentic Memory Demo

Traditional assistants send the entire conversation history to the LLM every turn — expensive, and quick to hit token limits. Ogcode's Agentic Session Memory extracts, stores, and retrieves only the context relevant to each query.

Benefit	Impact
~70% token savings	Drastically reduced API costs on long sessions
Infinite context	No practical limit on session length or codebase size
Higher accuracy	Only relevant memories are retrieved per query

How it works

Ogcode maintains a persistent Topic → Concept → Fact hierarchy with vector embeddings, plus a function-level Call Graph for codebase navigation. This knowledge graph survives across sessions, so your agent remembers your codebase structure, your conventions, and your past decisions.

Topic: "Ogcode Authentication"
  └─ Concept: "JWT Middleware"
     └─ Fact: "Token validation lives in internal/auth/jwt.go:47"
     └─ Fact: "Refresh tokens expire after 7 days (config: AUTH_REFRESH_TTL)"
  └─ Concept: "OAuth Flow"
     └─ Fact: "GitHub OAuth uses PKCE, implemented in internal/auth/oauth.go"

Enable it with:

export OGCODE_AGENTIC_MEMORY_MODE=true

Architecture

Ogcode is a single Go binary that embeds a SolidJS web UI and runs its own HTTP server.

┌─────────────┐     REST + SSE      ┌──────────────┐
│  Web UI     │ ◄────────────────► │  Go Server   │
│  (SolidJS)  │                    │  (port 9595) │
└─────────────┘                    └──────┬───────┘
                                          │
                    ┌─────────────────────┼─────────────────────┐
                    ▼                     ▼                     ▼
            ┌─────────────┐    ┌─────────────┐    ┌──────────────┐
            │ Agent Loop  │    │  SQLite DB  │    │ LLM Provider │
            │ (Claude,    │    │ (workspace  │    │ (Anthropic,  │
            │  GPT, etc.) │    │  + config)  │    │ OpenAI, ...) │
            └─────────────┘    └─────────────┘    └──────────────┘
                    │
                    ▼
            ┌─────────────┐    ┌─────────────┐    ┌──────────────┐
            │  Knowledge  │    │   Call      │    │   Search    │
            │   Graph     │    │   Graph     │    │   Bridge    │
            │  (Memory)   │    │ (Code Rel)  │    │  (Web/JS)   │
            └─────────────┘    └─────────────┘    └──────────────┘

Component	Responsibility
Agent Loop	Streaming LLM chat with tool execution (bash, read, write, edit, glob, grep, memory_recall, callgraph, deep_search)
Session Store	SQLite database for conversations, plans, tasks, and permissions
Git Worktrees	An isolated branch per task, so multiple agents work in parallel
Knowledge Graph	Persistent semantic memory with vector embeddings
Call Graph	Function-level code relationship tracking
Search Bridge	Playwright-based headless Chrome for web research

Roadmap

Advanced Task Planning & Parallel Execution — Enhanced plan decomposition with manual/automatic agent assignment to tasks.
AI Daily Standups — Voice-enabled meetings where agents report progress and discuss their work.
Ogland Integration — Connect external services (Slack, Email, Jira) for planning-phase use.
Agentic Deployment — End-to-end agentic deployment for major cloud providers, starting with AWS.

Community

Join the Ogcode community on Discord to ask questions, share feedback and feature ideas, and stay up to date with releases.

Join us on Discord →

If Ogcode is useful to you, starring the repo helps more developers discover it.

Contributing

Contributions are welcome — bug fixes, features, and documentation alike.

Fork the repository.
Create a feature branch: git checkout -b feature/my-improvement.
Make your changes and add tests where applicable.
Submit a pull request.

Please ensure your code follows the existing Go style and passes go test ./....

Security

Destructive operations require approval — The agent cannot write files or run shell commands without your explicit permission. Approve once, or always per tool.
Git worktree isolation — Each task runs in a separate git worktree, preventing accidental contamination of your working branch.
No data sent to third parties — Your codebase is analyzed locally. Only conversation text is sent to your chosen LLM provider.
Local-first — All data (sessions, plans, memory) is stored in local SQLite databases.

For security concerns, please open an issue or reach out on Discord.

License

MIT License — see LICENSE for details.

Made with care by the Ogcode team and contributors.

Star on GitHub · Discord · Docs

ogcode prasenjeet-symon

README

Ogcode

Context Engineering — the real differentiator

This is unique to Ogcode

Context engineering lets smaller models punch above their weight

Infinite Context — never hit a model's limit

Three wins from one idea

Plan Mode + Parallel PRs — ship features, not just suggestions

Plan once, ship the whole feature

How Ogcode breaks the feature down

Why this matters

How Ogcode Compares

Features

Table of Contents

Quick Start

Why Ogcode

Token Efficiency

How Ogcode saves tokens

Real session results

System Requirements

Installation

macOS / Linux

Windows

Go Install

Docker

Configuration

AI Provider

Ollama (local models — free, private)

Agentic Memory (optional)

Web Search (optional)

Usage

Build Mode (default)

Plan Mode

Custom port

Remote Deployment

Expose the port

Behind a reverse proxy with HTTPS

Docker Compose

Security considerations

Rich output

Agentic Session Memory

How it works

Architecture

Roadmap

Community

Contributing

Security

License