Product

Philotic Mobile Philotic Desktop Philotic Platform

Industries Enterprise Pricing Case studies

Philotic Platform

Own the stack you run AI on. Platform is the substrate behind every Philotic surface — and yours.

What Platform is

The kernel underneath every Philotic agent — yours and ours.

The same substrate that powers Philotic Mobile and Desktop. Exposed at a lower level for the people who need to see what their AI is doing, change how it does it, and own the weights, data, and perimeter it runs on.

Most users will never need it. They'll spin up agents in the apps and get on with their day. Platform is what's underneath when those agents run, and what you reach for when you've outgrown the apps — when you're building agents the apps don't ship, when an auditor wants every inference traced, or when your data can't leave a perimeter you control.

For

Power users, AI engineering leads, and the CTO offices owning the AI stack.

Replaces

A vendor lock-in to one model API, plus the four point-tools you bolted on around it.

Runs on

Our cloud, your VPC, on-prem, or air-gapped — same kernel, you choose the perimeter.

How to think about it

Run · See · Own. Three guarantees that show up at every layer of the stack.

01 · Run

Any model. Any agent. One substrate.

Frontier APIs, open weights, Philotic-1 variants, and the fine-tunes you make. One SDK, one router, one bill.

Jump to section ↓

02 · See

Every inference accounted for.

X-Ray ships with every response. Replay any agent run, audit any retrieval, see exactly what your AI did and why.

Jump to section ↓

03 · Own

Your weights, your data, your perimeter.

Bring your own keys. Hold your own KMS. Pin your residency. Deploy on your substrate. The default, not a SKU.

Jump to section ↓

01 · Run

Any model. Any agent. One substrate.

Platform speaks one wire protocol to everything below it: Philotic-1 variants, frontier model APIs, open-weights you self-host, and the fine-tunes you create. No bespoke clients. No second SDK. The router picks the right model for the request — or you pin one — and the rest of the stack stays the same.

Philotic-1 family — auto, fast, reason, auto-tune, sovereign
Frontier connectors — Claude, GPT, Gemini, with your contracts honored
Open weights — Llama, Qwen, DeepSeek, run on your hardware
Bring-your-own — point at any provider that speaks the wire
Routing rules you write, not ones we hide

Models

One catalog, every posture.

Philotic-1 is a portfolio of variants under one branded endpoint. Frontier and open-weights models you connect join the same surface, the same trace, the same bill.

Philotic-1 · auto

Picks the best variant per query. The default.

router

Philotic-1 · fast

Sub-second time-to-first-token at scale.

low-latency

Philotic-1 · reason

Long-context reasoning and structured output.

reasoning

Philotic-1 · auto-tune

Domain-adapted under operator approval.

fine-tune

Philotic-1 · private

Air-gappable build with customer-held keys.

on-prem

Frontier connectors

Claude, GPT, Gemini under your contracts.

connected

Open weights

Llama, Qwen, DeepSeek — your hardware, our trace.

self-hosted

Your fine-tunes

Adapters trained on your data, owned by you.

private

02 · See

Every inference accounted for.

X-Ray is shipped with every response. Not a sampled trace, not an opt-in feature, not a separate observability vendor — it's part of the response object. You can replay any agent run, audit any retrieval, and answer the question your auditor is going to ask: why did this model say what it said?

Per-inference trace

Chunks retrieved, model chosen, tools invoked, alignment events, latency, cost, stop reason.

Replay any run

Reconstruct the exact inputs, prompt, and context. Reproduce a result months after it ran.

Audit log export

Stream traces to your warehouse, your SIEM, or a flat file. We don’t hold the source of truth hostage.

Confidence & alignment

Confidence scores, citation pressure, policy outcomes — surfaced on every response.

03 · Own

Your weights, your data, your perimeter.

The reason most companies can't put their best work into a frontier API is that they'd be handing over the asset they're building. Platform inverts that. The keys are yours. The fine-tunes are yours. The retrieval index is yours. We provide the kernel; you provide the perimeter.

Bring-your-own model keys (Anthropic, OpenAI, Google, your own)
Customer-held KMS encryption — Philotic never sees keys unwrapped
Fine-tunes live in your tenancy. Not shared. Not extracted.
Residency pinning at retrieval and inference time
Export anything, anytime — traces, fine-tunes, indexes

Deploy

One kernel, four perimeters. Move along the spectrum as the work demands. Same SDK, same trace, same agents.

01 · Cloud

Philotic-managed.

Fastest path. Multi-tenant kernel, per-tenant data and keys. The default for Pro and Team.

Self-serve · day one

02 · VPC

Your cloud account.

Single-tenant kernel inside your AWS, GCP, or Azure account. We deploy and operate; you own the substrate.

Power & Enterprise · weeks

03 · On-prem

Your data center.

The same build, running on metal you control. Your hardware, your network, our kernel.

Enterprise · 6–8 weeks

04 · Air-gapped

No outbound, ever.

Disconnected build for classified, regulated, or sensitive deployments. Updates ship by signed bundle.

Enterprise · custom

Developer

An SDK that moves at the speed of your team.

The wire protocol is OpenAI-compatible — change the base URL and most clients keep working. Every response carries the trace alongside the standard usage and content. If you're evaluating whether to leave a frontier API for a substrate you control, the migration is hours, not weeks.

pythontypescriptgocurlopenai-compat

chat.pyphilotic ≥ 0.4

from philotic import Philotic
 
client = Philotic(api_key=os.environ["PHILOTIC_API_KEY"])
 
response = client.chat.completions.create(
    model="philotic-1",
    messages=[
        {"role": "user", "content": "What changed in the Caliber RFP?"},
    ],
)
 
# every response carries the trace
print(response.choices[0].message.content)
print(response.trace.confidence, response.trace.chunks)

Agents

Agents you build on Platform run on every Philotic surface.

Compose an agent in the SDK. It shows up in Philotic Mobile, Philotic Desktop, and on the web — gated by the policies and approvals you set. Same trace, same identity, same usage meter, no second deploy.

Process model — agents are processes, not chat threads
Approval policy — pin which actions require a human signature
Connector library — GSuite, Slack, GitHub, Notion, your own
Long-running missions, not turn-by-turn
Pre-built personas — Planner, Email, Sales — extend or replace

Security & compliance

The boring parts your security team will check first.

Audited

SOC 2 Type II

Compliant

GDPR · UAE PDPL

Registered

ADSIC ND 133684

Default

BYO-KMS

On request

Customer DPA

On request

Pen-test reports

Build on the substrate you can audit. Move when the work demands. Own what you build.

Talk to the Platform team Try the SDK on your free account