Integrations

Where your model runs is your call.

Three native providers. Same agent platform. Pick the one that matches your compliance posture — switch later without changing a line of agent config.

Local Self-Hosted

Default

Air-gappable, fully sovereign. Your models, your hardware, zero external dependencies.

Configuration

# Default — no config needed
OLLAMA_URL=http://localhost:11434

# Pull a model and go
ollama pull qwen3:235b-a22b

Available models: Qwen3 235B, Llama 3.3 70B, Mistral Large, Kimi K2.6, DeepSeek V3, any GGUF model

Compliance posture

  • ›Data never leaves your network
  • ›No vendor cloud dependency
  • ›Full air-gap capable
  • ›No training on your data — models run locally

Best for

Teams that need complete data sovereignty, defense/government, or want the lowest long-term cost.

OpenRouter

Fastest setup

Access 200+ models from one API key. Cheapest path to production — pay per token, no hardware.

Configuration

PROVIDER=openrouter
OPENROUTER_API_KEY=sk-or-...

Available models: Kimi K2.6, GPT-4o, Claude Sonnet, Qwen3, Llama 3.3, Gemini 2.5 Pro, 200+ more

Compliance posture

  • ›Data routes through OpenRouter (US)
  • ›No BAA available
  • ›OpenRouter retains logs per their policy
  • ›Best for non-regulated workloads

Best for

Startups, non-regulated teams, or anyone who wants to test multiple models before committing to hardware.

AWS Bedrock

HIPAA / FedRAMP

Native Bedrock integration. Claude, Llama, and Mistral on AWS with HIPAA BAA and FedRAMP path.

Configuration

PROVIDER=bedrock
AWS_REGION=us-east-1
# Uses your IAM role / instance profile

Available models: Claude Opus, Claude Sonnet, Llama 3.3, Mistral Large, Cohere Command R+

Compliance posture

  • ›Data stays in your AWS account
  • ›AWS BAA covers Bedrock inference
  • ›FedRAMP Moderate path available
  • ›No training on your data — AWS Bedrock policy

Best for

HealthTech, FinTech, LegalTech, Government — any team that needs a BAA or FedRAMP.

Custom integrations

Connect any tool, not just the ones we ship.

18 integration packs are pre-built (HubSpot, Salesforce, Stripe, Zendesk, etc.). For everything else — including your homegrown internal tools — Inhouse generates the API wrapper itself.

📋

Paste API docs, get a working integration

Drop in an OpenAPI spec, Postman collection, or a markdown description of the endpoints. Inhouse parses it, generates tool bindings, and your agents start calling it within a minute.

# Example formats accepted
https://docs.acme.com/openapi.yaml
postman_collection_v2.json
endpoints.md
🔧

Self-heals when APIs change

Field renamed? New required header? Endpoint deprecated? Agents inspect the response body, infer the correction, and retry — usually without a redeploy.

⚠API returned 400: "field renamed company_name → org_name"
✓Wrapper auto-corrected · retry succeeded · 1.3s total
🔐

Every auth method, native

OAuth (with your own client app), bearer tokens, basic auth, API keys, mTLS. Credentials encrypted at rest with AES-256-GCM. Per-agent scoping if you need it.

OAuth 2.0 + PKCE Bearer API key Basic mTLS
⚡

Production-aware

Respects rate limits and retry-after headers. Exponential backoff on 5xx. Circuit breaker if an upstream is degraded. All actions logged in your audit trail.

  • ✓Rate-limit aware (429 retry-after honored)
  • ✓Exponential backoff · circuit breaker
  • ✓Every call logged with timing + outcome

Used in production for HubSpot, Salesforce, Stripe, Zendesk, Linear, Asana, and several customers' homegrown internal systems.

One platform, any provider

Your agents, roles, tools, and workflows stay identical regardless of which provider you choose. Switching is a config change — not a migration.

1

Pick your provider

Local Ollama for sovereignty, OpenRouter for speed, Bedrock for compliance.

2

Set the env var

One line: PROVIDER=bedrock. That's it. No code changes, no agent reconfiguration.

3

Agents come online

Same roles, same tools, same personality. Different inference backend. Switch anytime.

Not sure which provider fits?

15-min call. We'll match your compliance requirements, budget, and workload to the right inference path.

Book a 15-min demo →