01 · Decide
See what's wrong.
Deterministic findings from your repo — risky files, complexity, debt, untested high-risk code. Not guesses.
Local-first · Open source · MIT — No account · Any local git repo · Never executes your code
Reikon is the local-first seat between you and your coding agents. It finds what's actually wrong, you dispatch your own agent to fix it, and the fix comes back as a diff you review — accept, or send back with a reason.
Deterministic findings from your repo — risky files, complexity, debt, untested high-risk code. Not guesses.
Hand the hunk and your reasoning to your own coding agent over MCP. Request fix — it works where it already runs.
The diff comes back linked to the finding that started it. Accept it — or send it back with a reason.
01 / The loop
The whole product. Work starts from what's measurably wrong — and you review every diff that returns. The dispatch ledger tracks what's out, what's back, and what's pending — across every open repo.
Deterministic findings from your repo — risky files, complexity, debt, untested high-risk code — plus agent flags gated by Reikon's own data, so they're measurably risky, not guesses.
From any finding: reply, resolve, or Request fix — handing the hunk and your reasoning to your own coding agent over MCP.
It reads the flag, does the work where it already runs, and the change comes back as an uncommitted diff in Reikon, linked to the finding that started it.
Accept it, or send it back with your reason — which rides along to the agent on the next pass. Fire-once becomes a real iterate loop.
02 / The seat
Each does one thing, then stops. Reikon is where you decide what's worth doing — and judge what comes back.
Verdict on a change someone already wrote. Then it stops.
Runs a pile of work in parallel — work you brought it.
Numbers you read once and a report you close.
The better your agents get, the more the bottleneck becomes the judgment directing them. That's the seat Reikon is built around.
03 / The wheel
Deciding about code is a loop — set the intent, then judge the act, the result, and the form. All four, shipped.
04 / Private by architecture
No telemetry. No account. No API key. All analysis runs locally — and the only thing that ever leaves is what you choose: a report you export, or prompts to an AI tool you drive yourself.
Privacy isn't a tier you upgrade to. It's the architecture — and it's why a hosted PR bot structurally can't be Reikon.
05 / The evidence
Every finding stands on deterministic intelligence — documented, reproducible, no black box.
06 / Built on MCP
Your agent already speaks MCP. Reikon answers in one round-trip — read, write, and the context that makes it sharper.
One question, one answer — not 15–30 calls rebuilding context by hand.
Your stated goals, and whether you've already decided this exact thing before — so the agent is sharper, not just faster.
Not just reading: flag a hunk, resolve, annotate — your agent acts through the same protocol it reads from.
Decisions and threads survive edits and commits; notes re-anchor to the code as it changes, and resolved/dismissed history stays — the context a hosted bot throws away at merge.
Any MCP-compatible coding agent. Optional AI is keyless — it drives a CLI you already have, never required.
The reviewer's verdict, the analyzer's numbers, the agent's labor — all of it is in service of one decision, and the decision is yours.
07 / Get Reikon
Any MCP-compatible coding agent works the loop. Optional AI is keyless — it drives a CLI you already have (Claude Code, Codex, or Gemini), never a cloud service of ours, never required.
08 / What's next
What we're actively working on — the full roadmap, tiered Now / Next / Exploring, lives on the changelog. Like everything in Reikon, it stays local-git only.
You decide and judge; the agent does the work; your code stays on your machine. Free and open source.