Agents
Our crew, and how it works.
We run a crew of AI agents for our own work: researching, writing, building, checking. This page explains what each one does and how they work together. Read as much as you like.
The agents themselves are not a free download. Their full files come with a crew setup, fitted to your work, from R1,200. That price is small on purpose. It keeps a small company going, and it pays for what we give away.
Our personal settings stay on our own machines. Three of our agents stay private altogether, because their work is personal or belongs to someone else, so they are not in this catalog.
The thing worth stealing, though, is the method, not any single agent. No agent grades its own work. The writer never checks the writing. The builder never signs off the build. A separate agent with different instructions tries to break it, and a final gate decides whether it ships. It's slower, and it catches what a single clever agent misses.
Later, once the community has checked enough of our scripture cards, we'll host a scripture agent you can ask questions directly. It will only answer from checked material, and it will show you where every answer came from.
SAFE ON BOTH SIDES
What we hand over, you can check.
This page tells you what each agent does. The agent files themselves reach you only with a crew setup. This site never hands one out, and neither does anyone else on our behalf.
Every file we hand over carries a checksum: a short fingerprint of the file. If the file on your machine has the same fingerprint, it is exactly what we gave you, and nobody changed it on the way.
Before you run any agent, ours included, scan it. We recommend NVIDIA's free SkillSpector scanner, with Cisco's free skill-scanner as a second opinion. Both are built for AI agent skills and look for things like hidden instructions and data theft. A clean scan lowers the risk but does not prove a file is safe, which is why the checksum and our record of every change matter too.
This protects both sides. You know what you are running, and we can show exactly what we handed over. It is the same thing our vision asks of everything we do: be open about how we work, be honest about it, and keep a record of every change.
THE METHOD
Nobody marks their own homework.
Every job moves through stages: research, draft, build, check, gate. Each stage leaves a written record. If the record is missing, the work goes back. The agent that made something is never the one that approves it. You can copy this setup for your own team. It works for people too.
The catalog
These agents draft, check and organise. They do not replace reading the text, and nothing they produce is published without a human reading it.
Public version. Our personal settings and logs stay private.
Scripture study
Text Before Take
PlannedScores a draft or page for textual fidelity, accuracy and trust risk before you post it. It never rewrites.
- For
- anyone teaching scripture publicly, from a church page to a podcast.
- You get
- a scored audit report with the specific lines that worry it and why.
Plumbline
PlannedTakes a list of flagged claims and gives each one a verdict (confirmed, corrected, partly, unresolved) with the evidence, written to a file as it goes.
- For
- writers and study groups who have claims they are not sure of.
- You get
- a verdict log that survives a cut-off, so you can rerun and resume.
Study Drafter
PlannedDrafts short study cards (historical context, common misconceptions, apologetics angle). Drafts only.
- For
- apologetics students and group leaders who want a first draft to check, not an answer to trust.
- You get
- a draft file marked status: draft, with its sources listed and open questions flagged for you.
Passage Research Packet
PlannedGathers the original-language words, context, named sources and claims to avoid for a passage, and never writes the teaching for you.
- For
- preachers and teachers who want their prep time to be writing time.
- You get
- one research packet per passage.
Writing
Peki
PlannedOne writer, in the register you choose up front: apologetics or commercial. Drafts only, and it never grades its own work.
- For
- small ministries and small businesses that write a lot.
- You get
- a draft with its register stated at the top and its uncertain points flagged. Pair it with Text Before Take for the check.
Series Pipeline Board
PlannedShows where each piece in a numbered writing series stands and which three to do next.
- For
- anyone writing a long series (50 devotionals, a course).
- You get
- a status board built from the files you already have.
Publish Pack
PlannedTurns one finished, checked piece into a caption, a carousel outline and design blocks, using your own words.
- For
- solo publishers.
- You get
- derivative formats, never new teaching.
Research
Pathfinder
PlannedAnswers a real question with a dated report: sources, anything unsourced marked as such, and a recommendation.
- For
- founders and leaders deciding something.
- You get
- one research report with links.
Steward
PlannedSizes a big multi-agent job against your plan's usage window before you start, and tallies where the tokens went afterwards. It never trades accuracy for tokens.
- For
- anyone running more than a few agents at once on a limited plan.
- You get
- a run plan before, a token tally and leak list after.
Finish or Ask
PlannedChecks whether a long job fits before starting, and asks you rather than handing back half the work.
- For
- everyone who has been burned by a half-finished bulk job.
- You get
- a size check and a clear question.
Build and verify
Reading UX Reviewer
PlannedReviews screens that carry dense text and data, and proposes concrete fixes: collapse by default, scoped search, work done ahead of time.
- For
- app builders whose content keeps growing.
- You get
- specific interface proposals, checked against the real size of the data.
Builder
PlannedImplements a spec and returns working code with build evidence. It never verifies its own work.
- For
- solo developers pairing a builder with a separate checker.
- You get
- code plus a build-evidence file.
Proof
PlannedReproduces, runs and tries to break what the builder claims, and returns pass or fail per claim. It has no edit tool, by design.
- For
- anyone who has shipped something "done" that was not.
- You get
- a verification report with a verdict per claim.
Designer
PlannedWrites the design spec (colour, type, layout, assets) that a builder then implements. It does not build.
- For
- small teams with no designer.
- You get
- a design spec file.
Gate
PlannedReads the whole chain of files from a workflow and returns one verdict: approved, revise or blocked. Reviews only; never builds.
- For
- anyone running a multi-agent workflow who wants a final independent check.
- You get
- a gate report with the verdict and reasons.
Premium Site Workflow
PlannedA repeatable path to a high-end motion website: brief, references, locked design system, section-by-section build, screenshot critique.
- For
- people building their own site.
- You get
- a staged workflow.
Business
Funding Finder
PlannedFinds South African grants and funding that fit your business and drafts application content. It never submits anything.
- For
- South African founders and NPOs.
- You get
- a dated list of opportunities and draft application text.
Outreach Follow-up
PlannedTracks who you contacted and drafts the follow-up. It never sends.
- For
- founders chasing investors, funders or clients.
- You get
- a due list and draft messages in your tracker and your mailbox's drafts folder.
Delivery Lead
PlannedTakes one client outcome through a fixed ladder (ground truth, stress test, risk register, plan, client communication, client enablement last) and does not stop until it is proven.
- For
- freelancers and small agencies delivering paid builds.
- You get
- a phase-by-phase project file with a gate at each rung.
Channel and creator
Channel Hunter
PlannedFinds channels that are really winning in your niche and proves it with numbers, then picks the one to study.
- For
- new video creators.
- You get
- a dated channel report with views per video, cadence and growth. Revenue figures are estimates and labelled.
Decoder
PlannedDissects winning and losing videos into a one-page formula and a fill-in script template.
- For
- creators who want a repeatable structure.
- You get
- a formula page, a script template and a Shorts cut.
PROMPT INDEX
The prompts behind the work.
PlannedA searchable index of the prompts we actually use, for scripture study, research, writing and building. Copy one, change it, make it yours.
Coming soon.
COMING LATER
Ask it anything. It'll show its sources.
PlannedA hosted scripture agent that answers only from cards our community has checked. Code will pull every verse, word for word. Every answer will be cited, and every correction public. It comes after the checked library is big enough to deserve it, and not before.
Help build the library