Manna Code

FEATURES

A full agent on your desktop — nothing rented, nothing leaked.

Use your own OpenAI or Grok subscription — a Grok 4.6 plan gets infinite context because threads never compact. Headroom compression is built in, so oversized tool output stays lean without losing the original. Full computer automation and background workers handle the long jobs. Import or stay in sync with CLAUDE.md and agent.md from Claude Code, Codex Desktop, Antigravity, Cursor, Grok Build, and other harnesses — and bring themes, including VS Code. ~3x fewer tokens in a measured benchmark, on your keys, your models, your machine.

YOUR KEYS · YOUR SUBSCRIPTION

Bring your own OpenAI or Grok subscription.

No credits required. Plug in the plan you already pay for. Pair a new Grok 4.6 subscription with Manna Code and threads never compact — infinite context on the plan you already have.

GROK 4.6

Infinite context on your subscription

Use the Grok 4.6 plan you already have. Threads never compact — day-one details stay verbatim on day thirty.

OPENAI

Or bring your OpenAI plan

Same agent, same local-first setup. Your OpenAI subscription, your keys, your machine.

CONTEXT & MEMORY

“Infinite” context

Never compact again. Chat threads have unlimited length and no lost history — the conversation you had on day one is still there, verbatim, on day thirty. Pair it with a Grok 4.6 subscription and you get infinite context on the plan you already have.

Headroom compression, built in

Large tool results are compressed automatically so the live context stays lean. The original stays on your machine — retrieve the exact lines, values, or errors when you need them. No plugin, no extra setup.

Context browser

Always know exactly what's in your context. Browse it live, then make targeted tweaks — drop what's stale, pin what matters — to minimize or maximize token usage.

~3x fewer tokens — measured

In a 10-run single-prompt code-audit benchmark, Manna Code moved 149K tokens per run versus Claude Code's 439K. Same Opus 4.8 model and effort; both graded runs scored 11/11.

AGENT MODES

Analyst mode

Work with files, documents, spreadsheets, data, and the web. Attach your corporate database to give the analyst company knowledge — raw data in, Python, charts, and answers out, streamed step by step.

Code mode

A planning-first coding loop with parallel tool execution, subagents, todo tracking, and mid-task check-ins. It reads your codebase, edits files, runs tests, and shows its work.

Pause and steer

Pause the agent mid-response and inject any thought you'd like — a correction, a constraint, a better idea. It resumes with your input in context, no restart required.

AUTOMATION

Full computer automation

The agent doesn't stop at files. It can operate your desktop — click, type, open apps, fill forms, and drive a browser — so real workflows get done, not just described.

Background workers

Hand off long-running work and keep going. Background workers run the jobs that take hours — overnight refactors, multi-step automations, anything you don't want to babysit — while you stay in the conversation.

From browser to desktop

A single job can browse the web, operate apps, and write the result back to disk. Automation is part of the agent, not a sidecar you have to wire up.

MODELS & ORCHESTRATION

Bring your own keys — or your subscription

Use the OpenAI or Grok subscription you already pay for — including Grok 4.6 — or plug in local models, Fireworks.ai, Together.ai, OpenRouter, or any OpenAI- or Anthropic-SDK-compatible endpoint. You hold the keys; we never see them.

Build your own orchestration

Assign models to the roles you define: main orchestrator, vision, PDF, subagent, code, analyst. Each function runs on the model you chose for it — the session reports what it spent.

Local + frontier, together

Route routine work to local models and hard problems to stronger API models through your own custom orchestration. The bill drops without giving up frontier quality where it counts.

WORKSPACE

Organized by project

Group and sort chats by project, folder, or repo. Every thread stays where the work lives, with its full history intact.

File and git browser

A built-in file and git browser keeps the working tree, diffs, and history one click away — no switching to another tool mid-session.

FROM OTHER HARNESSES

Import or sync CLAUDE.md and agent.md

Bring the instruction files you already keep in Claude Code, Codex Desktop, Antigravity, Cursor, Grok Build, and other agentic harnesses. Import them once, or stay in sync so Manna Code and your other tools share the same project rules.

Import themes — including VS Code

Bring the look you already work in. Import themes from other harnesses, including VS Code, instead of starting from a blank palette.

Keep both tools if you want

Switching harnesses shouldn't mean rewriting your setup. Your CLAUDE.md, agent.md, and theme come with you — then stay current if you keep using Claude Code, Cursor, or the rest.

SECURITY POSTURE

Keys never leave the OS vault

API keys and database secrets are stored in the Windows Credential Manager, never in config files. Config keeps a reference, not the secret.

Data stays on the machine

Chats, events, attachments, and UI state live in a local database under your user profile. No cloud sync, no shared drive, no vendor backend in the loop.

Your database, on your terms

Attach your corporate database and the analyst answers with company knowledge — but SQL access is off until you configure a connection; the model can't even see the tool. Snowflake today; Postgres, Trino, and Databricks planned.

Ready to put an agent on your desk?

Get Manna Code