You choose the endpoint
Connect a self-hosted model service, compatible gateway, or internal API instead of hard-coding one provider into the client.
Checking sign-in status
01CONTROLLED RUNTIME / SELF-HOSTED 08
Run AI coding work in a controlled environment
When code cannot casually enter a third-party SaaS, Cocode provides a client entry point for self-hosted endpoints, private gateways, and controlled workspaces.
Capabilities and limits follow the current release
02Direct answer
Start with what it is.
A self-hosted AI coding agent runs the Host, model endpoint, or gateway in an environment controlled by your team. It increases infrastructure and data-path control but does not automatically provide encryption, access control, audit, backup, isolation, or security updates.
- Control surface
- Host, endpoints, network, logs, and runtime environment
- Deployment
- Local machines, private servers, VPCs, or controlled clusters
- Security conclusion
- Self-hosted does not automatically mean secure
- Operations
- Identity, secrets, TLS, backup, upgrades, monitoring, and recovery belong to the team
Workspace boundaries stay clear
Files, commands, network actions, and model requests have visible boundaries that are easier to review in a team environment.
Keep the full client experience
A private endpoint does not mean falling back to scripts. GUI, TUI, sessions, diffs, and tests remain one workflow.
02Why it matters
Self-hosted is about boundaries, not a label
Cocode treats self-hosting as an engineering boundary: know where files live, where requests go, which actions need confirmation, and how results are recorded.
↗
Clear boundaries make real development work easier to trust.
Self-hosted services and private gateways are first-class entry points
Choose an internal service, private gateway, or compatible API based on security and cost requirements. The client connects it to the same task context.
Review on desktop, execute over SSH
A developer can review task context, files, and diffs in the GUI, then continue execution from a controlled server through TUI or SSH.
05Working flow
Move from setup to verification with a clear result at every step.
This is an executable, reviewable path through a real repository—not an abstract feature list.
- 01
Single-machine topology
Run the client, Host, and compatible model endpoint on one controlled machine for experiments or personal use.
- 02
Team gateway topology
Connect developer clients to one gateway that centralizes models, credentials, quota, and audit controls.
- 03
Isolated-environment topology
Keep workspace, runtime, and model service inside a restricted network with explicit dependency and release paths.
- 04
Validate protocols before scaling
Test identity, Chat/Responses, streaming, tool use, timeouts, and error semantics independently.
06Use cases
Decide by the work you actually need to finish.
A controlled enterprise model entry point
Centralize providers, routing, credentials, and policy while clients connect to one approved endpoint.
Sensitive-code environments
Place workspace, network, and model access inside a defined trust domain and audit scope.
Replaceable models and gateways
Switch model services within a verified compatibility boundary and reduce client lock-in.
07Before you start
Runtime requirements and compatibility boundaries
- TLS, identity, key rotation, and least privilege
- Explicit egress, log redaction, data retention, and backup policy
- Verified endpoint compatibility with the required Chat Completions or Responses protocol
- Health checks, capacity, timeouts, failure recovery, and upgrade rollback
- Team ownership for configuration, logs, and high-risk approvals
08Try it
Minimum client-side check
A successful diagnostic only proves the client environment; endpoint protocol, security, and capacity need separate verification.
cocode://workspaceREADY
cocode doctor
cocode09Facts and boundaries
Self-hosting transfers control and responsibility together.
A running deployment proves availability at one moment, not security, resilience, or compliance.
| Capability | Status | Conditions and evidence scope |
|---|---|---|
| Infrastructure control | Team-owned | The deployer manages compute, network, storage, images, and runtime configuration. |
| Model location | Architecture-dependent | A self-hosted Host can still call an external model API; inference must be controlled separately. |
| Security controls | Must be configured | TLS, authentication, least privilege, audit, and isolation do not appear automatically. |
| Recovery | Must be verified | Backups, upgrades, rollback, health checks, and recovery exercises are operational responsibilities. |
10Cocode perspective
Self-hosted is a responsibility model, not a security label.
Cocode evaluates identity, secrets, network, logs, backup, upgrades, and rollback separately. Deployment control becomes a security benefit only after those controls are configured and verified.
- Maintained by
- Cocode Agency
- Last reviewed
- 2026-08-25
- Scope
- Self-hosted Hosts, model endpoints, network, security, and recovery responsibility
- Verification
- Maintained against current deployment boundaries, runtime requirements, and production release checks
06FAQ
Before you start, make the boundaries clear.
Is Cocode itself a self-hosted model service?
No. Cocode is the client and workflow layer. It supports self-hosted model endpoints and private gateways; whether the model service is self-hosted depends on your deployment.
Who is this for?
It fits developers, platform teams, and engineering organizations that need control over code boundaries, model request paths, network permissions, and task context.
Can I use only a local model?
If the model endpoint matches the protocol and capabilities supported by the current client, a local or internal model can be used as the entry point.
Does a self-hosted Host mean the model also runs locally?
Not necessarily. A controlled Host can still call a cloud model API, so inference location must be checked separately.
What is the minimum security baseline for self-hosting?
Cover TLS, authentication, secret storage, least privilege, network restrictions, logs, backups, upgrades, and a verified rollback path.