Bring your own keys
Your money goes to the model providers. Nakami does not host inference and does not resell tokens. Keys sit in the OS keychain, never in files.
- Anthropic
- OpenAI
- xAI
local-first · BYOK · source-available
Nakami is a free agentic IDE that plans your work into tickets, dispatches Claude, GPT, Gemini, or Grok on your own API keys, and ends every ticket in a diff you review. Everything stays on your machine.
Coming soon
Nakami is still in development. The download and the source open at launch.
runs on your keys —AnthropicOpenAIGooglexAI
trust
Three commitments, each verifiable — not marketing claims.
Your money goes to the model providers. Nakami does not host inference and does not resell tokens. Keys sit in the OS keychain, never in files.
Nothing leaves your machine. 中身 (nakami) means the contents — what's inside. The board, the keys, the logs, and the diffs stay contained within your computer. Cloud, when it exists, never gates a local feature.
The source opens for reading at launch. There is no telemetry to audit — nothing is collected.
workflow
The loop is the product: the planner writes tickets, agents work them, verification gates completion, and you review every diff.
Write an epic as a pitch. The planner proposes tickets with goals, checkable criteria, and dependencies. You approve before anything is written to the board.
The agent cannot declare a ticket done. It runs every acceptance criterion, records the evidence, and only then may the ticket reach review. A human reads the diff.
The log drain catches a stack trace. One click hands the crash and surrounding logs to the planner. A defect ticket is drafted, ready for an agent to pick up.
compare
Categories, not competitors. Every cell is verifiable.
| Question | Nakami | Metered cloud IDEs | Chat-panel agents |
|---|---|---|---|
| Who holds the keys | You. Keys live in the OS keychain, never in files. | The vendor. You buy inference from them. | Typically the vendor; BYOK is sometimes an option. |
| Where code and tickets live | Your machine. Tickets are markdown in the repo. | The vendor's cloud. | The chat transcript. There is no native ticket file. |
| Unit of work | A ticket: one agent, one branch, one reviewable diff. | A cloud agent run or composer session. | A conversation. |
| What done means | A human-reviewed diff. Only a human moves the ticket to done. | The run finishes. Human review is a separate step, if it exists. | The model stops generating. |
| Cost visibility | Input/output tokens and computed cost, rolled up per ticket in the Run Record. | Credits, usage pools, or a monthly seat. | Plan- or message-level; not rolled up to a ticket. |
pricing
local
$0
The desktop app is free. Every local feature stays free — cloud never gates it. Model usage is billed by your providers, at their rates, visible per ticket.
organizations
coming
Shared boards, org keys, and hosted runners are a later paid layer. They do not exist yet.