Forge gives you API access to a working, epistemically-labelled knowledge corpus — plus the reasoning layer behind CHIKARA — so you can build honest software instead of starting from a blank prompt.
Where Forge actually stands
We would rather tell you exactly what works today than promise something polished and make you find the gaps yourself.
524 entries across seven corpora — every answer carries an IRA confidence label (Known / Inferred / Hypothetical) so you know how much to trust it.
Register, log in, and query the corpus programmatically today. Session-based auth — no waitlist gatekeeping for this tier.
Cache-mode works with zero setup. Optional backends — Ollama (local), OpenRouter, or a CHIKARA endpoint — when you want coverage beyond the corpus. Not required to start.
What you are querying
Every entry was written and reviewed for this platform, then run through IRA before it shipped. Seven named corpora ship today.
How it works
No SDK required for early access — plain HTTP, JSON in, JSON out.
Create a builder account and get a session token.
POST /api/auth/register
{ "name", "email", "password" }
Ask in plain language, optionally scoped to a corpus key.
POST /api/query
{ "query": "...", "corpus": "software_engineering" }
Every response includes an IRA label and sources — never a bare string alone.
{ "ira_label": "KNOWN",
"confidence": 0.97,
"sources": [...] }
The reasoning layer
Forge uses the same Immaculate Reasoning Atom (IRA) engine that grades confidence across CHIKARA products. You get a classification — so your app can decide what to trust, what to flag, and what to reject.
Open the full factor weights, risk zones, and interactive demo on the metrics reference page.
Open metrics reference →Sibling surface
A separate FastAPI surface: Eighteen Factors weighted risk (0–100) and IRA claim classification. Not a substitute for the 524-entry corpus — a companion for high-stakes human context.
POST factor scores. Receive normalised risk, zone (GREEN / YELLOW / RED), and top contributors. Ubuntu/Dignity remains the heaviest weight.
POST /score
{ "context_label", "factor_scores" }
POST a claim and probability. Receive KNOWN / INFERRED / HYPOTHETICAL / UNVALIDATED.
POST /ira/classify
{ "claim", "probability" }
Scoring + optional IRA + Four Prerequisites warning when dignity and honor both run high.
POST /assess
→ zone + IRA + warning
Try the metrics demo (client-side) →
Framework credits: Eighteen Factors — Rodney Vianté Keliana; IRA — Rodney Manyakaidze. Deploy the scoring API from its own package when you need those endpoints live.
What is next
524-entry verified corpus, builder accounts, IRA-labelled query responses.
Point optional queries at Ollama, OpenRouter, or a CHIKARA endpoint. Default remains cache-only so builders are never blocked on a paid key.
Query history, per-domain tracking, multi-user builder accounts.
License the IRA reasoning layer into third-party products, not only CHIKARA's.
Access
Early access is open — real builders finding real gaps beat a polished waitlist page.
Questions
Yes. Register, authenticate, and query a real 524-entry corpus today. Optional live-model backends are configuration, not a blocker.
You get an honest lower-confidence label rather than a fabricated fact. With Ollama, OpenRouter, or a CHIKARA endpoint configured, routing is available; otherwise the corpus answer (or honest low confidence) is what you get.
The Immaculate Reasoning Atom — CHIKARA's confidence grading. See the metrics reference for bands, Eighteen Factors, and a live demo.
No. Forge search hits the verified corpus. The Cultural Reasoning API scores situations (Eighteen Factors) and classifies claims (IRA). Companion tools — not one product pretending to be both.
Get started
Register a Forge account and query the corpus in minutes. Study the metrics when you need the scoring layer.