Overview
plan —
Sign out
Efficiency — this window30 days
—%
cache hit rate
—ms
avg cache hit
—ms
avg LLM call
—
avg items / request
—
calls avoided
Tokens saved—
—
—
—
$ saved
—
context held back
—
memory, net of tokens added
—
from cache
—
avoided vs whole store (separate)
Mistakes headed off—
—
—
known mistakes resurfaced
—
outdated instructions retired
—
mistakes recorded

Resurfaced = a mistake your agent recorded was put back in front of it when the same situation came up again. Retired = you saved a changed version of an instruction, so the old one was archived and your agent stopped seeing it. Record one with POST /api/authority/mistake.

Memorymanage
Total memories—
Total accesses—
Items injected—
Requests memory reached—
Last injection—
Conflicts flagged—
Security posturedetail
Perimeter active
Contradiction active
Cross-tenant shared
Abstention fail-closed
Integrity verified
Backups hourly
Threats screened100%
Blocked / suspects—
Cache & latency—
Requests this window—
Served from cache—
p50 latency—
p95 latency—
Response caching
—
Recent memoriesreload
Loading…
Connect in two linesdocs
# point your client at Iconia
client = OpenAI(
  base_url="https://www.iconia.tech/v1",
  api_key="—"
)
memories —
Export
Add a memory
Search your vault

Manual search here is free. Once activated, your agent gets this automatically, in context, on every request.

Bring your data
⇪
Drop a file, or click to browse
.json · .csv · .txt
Your memoriesreload
Loading…
Memory Audit

Contradiction report, coherence score, and near-duplicate check across your whole vault.

Memory revisions—
—
—
Skill constellation—
✶ brightness = level (grows with real use)size = depthcolour = source ◆ recorded mistakeinside each: "do" rules one side, "don't" rules the other
Abilities—

Loading…

Live posture — this tenantlearn more
Perimeter active
Intrusion trap armed
Contradiction active
Burn monitor baselined
Cross-tenant shared
Abstention fail-closed
Integrity verified
Backups hourly
Screening 100%
Threat feed
Requests screened100%
Blocked / suspects—
Honeypots served—

Full Security control room and alerts available on Builder and above.

Tokens saved
—
—
Requests served
—
this window
$ saved
—
at each model's real price · conservative
Monthly usage
Requests this month—
Plan allowance—
Savings split
Context held back—
Memory, net of tokens added—
From cache—
Avoided vs whole store (separate, not in total)—
Model calls avoided—
Sessions—

Each conversation your agents ran through Iconia, newest first.

—

Marketplace — specialistscoming soon

Adopt vetted domain expertise in one click. A specialist is a screened bundle of policies, playbooks, and knowledge — it becomes memory in your store instantly and your agents employ it by recalling it in context. Every pack clears the same contradiction and safety checks your own writes do.

Current plan—
Requests this month—
Allowance—
API key
Key —
Last used—
BYOK setup

Rotating revokes the old key immediately and shows the new one once. Treat it like a password.

Connect your tooltake the tour

Point the tool you already use at Iconia — memory, security, and caching switch on for every call. No SDK, no rewrite. Grab your ick_ key under API Keys.

# Cursor → Settings → Models → OpenAI API — custom base URL
Base URL:  https://www.iconia.tech/v1
API Key:   ick_your_key

Paste those two lines — every message your agent sends stores and recalls memory.

# Claude Code — two environment variables
ANTHROPIC_BASE_URL=https://www.iconia.tech/v1
ANTHROPIC_API_KEY=ick_your_key

Run claude as usual — it remembers across every session.

# Hermes — config.yaml (or OPENAI_BASE_URL / OPENAI_API_KEY)
model:
  provider: custom
  base_url: "https://www.iconia.tech/v1"
  api_key:  "ick_your_key"

Model-agnostic + self-hosted — your always-on agent gets governed, portable memory.

from openai import OpenAI

client = OpenAI(
  base_url="https://www.iconia.tech/v1",
  api_key="ick_your_key",
)

Drop-in for any OpenAI-compatible app or script.

Verify it works

A live check — Iconia stores a memory, then recalls it from a differently-worded question. No model call, no cost.

Troubleshooting
  • 401 / invalid key — use your ick_ key from API Keys, not your provider key.
  • Client ignores the base URL — some IDE agents pin their endpoint; if yours won't honor a custom base URL, reach out.
  • 429 / quota — your plan's request limit was hit; upgrade under Plan & Upgrade.
  • Nothing recalled — memory injects only when a request needs it; ask something related to what you stored.
Account
Email—
Plan—
Change email
Change password
Email preferences
Sign out
Danger zone

Deactivate revokes your keys and disables the account (recoverable via support). Delete permanently removes your account and all memory — irreversible.

Your model provider —

Your model calls run on your own provider account — OpenAI, Anthropic, OpenRouter, or any OpenAI-compatible gateway — so your provider bills you directly, at the lower amount Iconia makes possible. Add your key here once and every request uses it automatically (tools like Hermes and Cursor only need your Iconia key). Your key is encrypted at rest and never returned to the browser.

Switching models inside Hermes, Cursor or Claude Code? Pick OpenRouter — one key covers GPT, Claude, Gemini, Llama, Qwen and more, so any model you choose just works.