Troubleshooting

Last updated 2026-10-06·Reviewed against Prism v0.3.26

First: three checks that solve most cases

# 1. Is the proxy answering?
curl http://127.0.0.1:11434/health
# {"status":"ok"}

# 2. What version and service is on the port?
curl http://127.0.0.1:11434/
# {"status":"ok","service":"prism","version":"0.3.26"}

# 3. Does Prism see your models?
curl http://127.0.0.1:11434/v1/models

Port 11434 is already in use

Prism defaults to 11434, which is also Ollama's default port. If a local Ollama server is running, move Prism to another port.

# PowerShell
$env:PRISM_PORT = "11435"

# bash
export PRISM_PORT=11435

After changing the port, re-sync your agents from the Agents tab — the base URL written into every agent config is derived from the port, so existing entries would still point at the old one.

An agent still talks to its own vendor

Claude Code, Droid and a few others watch and rewrite their own config files, which can drop the Prism entries. Prism re-syncs its managed block on every startup.

401, 403 or unsupported-model errors

Because there is no inbound Ollama /api/chatroute, a client that only speaks Ollama's native API will fail with a 404 here. See Ollama Clients.

Requests do not appear in Stats

/v1/messages/count_tokens returns an Anthropic-shaped 404 by design, because Prism does not estimate token counts. Some clients log this as a warning — it is not an error that affects the conversation.

SearXNG will not start

The first start bootstraps an isolated Python environment, which needs a one-time download of roughly 80 MB.

Deleting the searxngfolder under Prism's config directory resets the instance; Prism re-bootstraps on the next start.

MCP problems

# Server health, including per-server state and messages
curl http://127.0.0.1:11434/mcp/status -H "Authorization: Bearer prism"

Remote HTTP and SSE servers stay connected; only stdio child processes are reaped after 300 idle seconds and restarted on the next call. An idle state is normal. See MCP Gateway.

Streaming looks stalled

Prism forwards each event as it translates it, so a stall almost always means the upstream is slow or a tool call is waiting. Check the Stats tab for live tokens per second while the request runs — if TPS is moving, the stream is alive.

Logs and debug mode

Prism writes to proxy.log in its config directory. With debug_logs enabled, it also captures full request and response bodies per translation, which is what to inspect when a format translation looks wrong.

# Watch the proxy log live (macOS and Linux)
tail -f "$HOME/Library/Application Support/prism/proxy.log"   # macOS
tail -f "$HOME/.config/prism/logs/proxy.log"                  # Linux

Debug logging is a toggle in the Admin UI and a key in config.json. Turn it off afterwards: it records full traffic to disk, including anything sensitive in a prompt.

Resetting Prism

Running without the tray

Prism has a headless mode for servers and containers. Run the binary with --serve to start the proxy without the tray or admin window. There are no other command-line flags; ports and behaviour are configured through PRISM_PORT, PRISM_HOST, PRISM_ADMIN_PORT and config.json.

Frequently asked