Connic

Stop self-hosting
your AI agents

Your agents work. The problem is your week — Kubernetes upgrades, Docker rebuilds, Grafana dashboards, on-call pages. Move to Connic and the infrastructure stack disappears, while your agent code stays exactly where it is.

Start FreeTalk to Sales
Read the migration guide
agents/invoice-processor.yaml
# Define the whole agent in one YAML file
name: invoice-processor
model: gemini/gemini-2.5-pro
temperature: 0.3
system_prompt: |
You are an expert accountant.
Extract every field from the invoice
and verify the totals add up.
tools:
  - documents.parse
  - documents.extract_entities
  - database.store_invoice
Three problems you stop owning

Hosting, scaling, observability — handled

Self-hosting means you build hosting, scaling, and observability — and stay on call for them. Connic ships all three so your team stops carrying infrastructure pagers.

Hosting

Skip all this

  • Kubernetes manifests
  • Docker builds
  • CI/CD pipelines
  • Cluster upgrades

git push is the deploy. No containers to babysit, no pipelines to debug, no cluster upgrades on the calendar.

Scaling

Skip all this

  • HPA configuration
  • Resource tuning
  • Capacity planning
  • Idle resource costs

Managed runtime scales with demand. You pay for runs and minutes — not idle clusters waiting for traffic.

Observability

Skip all this

  • ELK / Loki setup
  • Prometheus metrics
  • Grafana dashboards
  • Alert configuration

Full execution traces are on by default. Every LLM call, tool call, and timing shows up in the dashboard — no stack to wire up.

The migration

You aren't rewriting anything

Your Python tools stay the same. Your prompts stay the same. The YAML wraps what you already have — and the infrastructure layer goes in the trash.

What you keep
  • Python tools

    Same functions, same logic — drop them in tools/

  • Agent logic

    Prompts and orchestration translate to YAML and middleware

  • Git workflow

    Branches, PRs, code review — unchanged

  • External integrations

    APIs, databases, services — reachable from custom tools

What you delete
  • Kubernetes configs

    Manifests, Helm charts, kubectl runbooks

  • Docker setup

    Dockerfiles, image builds, registry plumbing

  • CI/CD pipelines

    GitHub Actions, Jenkins, deploy scripts

  • Monitoring stack

    Prometheus, Grafana, custom alert rules

Follow the quickstart to cut over your first agent.

Self-hosted vs Connic

What you manage yourself today, and what Connic takes off your plate

Self-hosted vs Connic
FeatureSelf-HostedConnic
DeploymentDocker/K8s + CI/CDgit push
RollbacksManualOne-click
ObservabilityDIY (Prometheus, Grafana)Built-in
Knowledge base (RAG)DIY (vector DB)Built-in
ConnectorsCustom per systemFirst-party catalog
A/B testingNot includedIncluded
Automated evaluation (Judges)Not includedIncluded
Human-in-the-loop approvalsNot includedIncluded
Budget alerts & limitsNot includedIncluded
Team permissions (RBAC)DIYBuilt-in
Secrets managementVault / env filesBuilt-in
Auto-recoveryDIYBuilt-in

A migration partner, not a docs page

Cut over with the same controls — and the same compliance posture — you have today

Dedicated infrastructure

Enterprise plans come with dedicated compute, custom SLAs, and an account manager who runs the migration with your team end to end.

Compliance continuity

DPA, EU AI Act support, SOC 2 Type II, ISO 27001, and HIPAA available on Enterprise — so legal does not have to re-approve a new vendor stack. Audit logs ship on every tier from day one.

Migrate at your own pace

Not ready for a full cutover? Use Bridge to keep services in your VPC while agents run on Connic. Move workloads one at a time, on your timeline.

Keep exploring

Composer SDK

Define agents in YAML.

Bridges

Reach private services, no inbound ports.

Observability

Trace every step, every token.

Connectors

Wire agents to your real systems.

Environments

Per-env config and secrets.

Testing

Assertions and judges, in CI.

Frequently Asked Questions

Most teams complete migration in under a day. The connic migrate CLI converts LangChain, LangGraph, and Google ADK projects automatically — agents become YAML, tools become Python files in tools/. Review MIGRATION_REPORT.md, run connic lint, and clean up anything flagged for manual attention. There's no big-bang cutover required: you can run both systems in parallel during transition.

Yes. Tool implementations, prompts, and business rules transfer directly. What changes is the configuration format (YAML instead of framework-specific code) and the deployment mechanism (git push instead of Docker/Kubernetes). The connic migrate CLI handles most of the conversion automatically.

Use Bridge. A self-hosted bridge agent runs inside your VPC and opens an outbound WebSocket tunnel to Connic — no inbound ports needed. Your agents can reach internal Postgres, S3, Kafka, and HTTP services through the bridge as if they were local.

Self-hosting costs include compute (clusters or VMs), DevOps engineering time, monitoring tooling, and on-call burden. Connic charges for runs and usage on managed infrastructure with no idle costs. Most teams see meaningful savings once they factor in the full cost of operating their own platform — see the pricing page for current tiers.

Minimal. Tool implementations are plain Python functions. YAML configs are declarative and portable. A few advanced primitives (StopProcessing, AbortTool) use lightweight Connic imports, but your core business logic stays standard.