ReadyBase Solo Guide
From install to an AI assistant that understands your codebase, in about five minutes. No team setup, no configuration files to hand-write.
1. Install
ReadyBase is a single static binary. No dependencies, no runtime. Pick your platform:
curl -fsSL https://promptforce.ai/rb-install.sh | bash
iwr https://promptforce.ai/rb-install.ps1 -OutFile rb-install.ps1; .\rb-install.ps1
github.com/PromptForcePrime/readybase-public/releases
60-day Pro trial, on the house
ReadyBase Pro activates automatically on first run, no key to paste, no network call. You get the full solo feature set for 60 days. When it's time, renew your license →. (Team features are a separate upgrade.)
2. Onboard your repo
One command prepares your codebase for AI-assisted development, it scans the repo, writes the guidance files your AI assistant reads, sets up the integration, and builds the intelligence index:
readybase go .
This generates everything: your CLAUDE.md, AGENTS.md, guard rules tuned to your high-dependency files, MCP config, and session hooks. Your AI assistant now starts every session already knowing the codebase.
Solo by default
The solo profile is the default, sensible settings for one developer, nothing to tune. Team and agent profiles come later, when you need them.
3. Read your score
See how ready your codebase is for AI on a 0-100 scale:
You get a grade plus a breakdown across nine signals, test coverage, documentation, dependency health, how many people understand the code, complexity, and more. The score is weighted by risk, not size, so a large well-owned repo isn't penalised and a fragile one can't hide behind a big line count. Every weak spot comes with the specific fix that would raise it. The score is a to-do list, not just a verdict.
Alongside readiness, ReadyBase tracks a second axis, coherence: how consistently the codebase follows its own conventions, and where AI-written code has drifted from them.
4. Navigate safely
Once onboarded, your AI assistant gets the instincts of a senior engineer automatically. Behind the scenes, ReadyBase exposes MCP tools your assistant calls at the right moment:
readybase_explore— start here for any task or stacktrace: the files to look at first.readybase_relate— what depends on this file? The blast radius, deps, and tests, before an edit is made.readybase_story— how did this file get this way? Recent commits, authors, churn, and bus-factor.
Those three lead a broader toolset (score, coverage, convention, ownership, and more) — your assistant picks the right one; you don't call them by hand.
Blast-radius warnings
When your AI is about to edit a heavily-relied-upon file, ReadyBase flags it, "this file has many dependents, high blast radius", so a small change stops causing big outages. You don't have to do anything; the warning just shows up.
5. Watch for drift
Every codebase has a house style, most of it unwritten. As AI writes more of your code, small deviations pile up: a different error-handling shape here, an off-convention name there. Each passes review; together they become vibe drift. ReadyBase learns your conventions from the code itself and flags where new work has strayed:
Coherence, not just correctness
Drift isn't a bug, the code works. It's output that doesn't fit how your team builds. Catching it in the moment keeps the codebase feeling like one author, not a committee of strangers, which is exactly what keeps it AI-ready.
6. Open the dashboard
Prefer a browser to a terminal? Launch the local dashboard:
Open localhost:8765 to see your score, gaps, hotspots, and (if you have several repos) the portfolio ranking, all running locally, nothing sent anywhere. See the dashboard tour →
7. Let an agent work unsupervised
Everything above is advisory: ReadyBase tells you about risk and you decide. If you want an agent making changes while you are not reading every diff, you need the guard to actually stop it. That is opt-in, per repo:
The list is derived, not hand-written: files land on it because many things depend on them, or because they are AI-authored with no second reviewer. With .readybase/constraints.json present, an agent editing one of those files is refused before the edit happens, not warned afterwards.
When that refusal is wrong, clear one file rather than switching the guard off:
Approvals are per-file and expire on their own, two hours by default. The verdict afterwards still shows why the file was protected alongside who cleared it, so an approved edit reads as risky and allowed rather than as a file that was never risky.
Turning it off is a decision, not a workaround
To suspend enforcement for a session, export READYBASE_AGENT_MODE=off. The hook announces it on every edit, because a guard that is silently disabled looks exactly like one that found nothing. Editing the constraints file instead changes the policy for everyone who clones the repo.
8. Know how far your repo can go
Not every repo should be handed to an unattended agent. The autonomy ladder puts a number on it:
Four levels. L0 draft, L1 report-only, L2 assisted fixes, L3 unattended. Each is gated on evidence rather than a checkbox: unresolved structural drift holds a repo at L0 no matter how good the score is, and L3 additionally requires live cost telemetry, because an agent running with nobody watching and no spend visibility is a runaway you cannot see.
The audit names the single gate blocking your next level, so there is always one thing to fix rather than a list to interpret.
What's next
- Privacy & local-first, why your code never leaves your machine.
- Readiness Certificate, the one-page verdict for your team lead.
- Pair with Prefex, cut what each token costs, too.