Fill-in-the-middle (FIM) completion for editor inline suggestions. Given the text before and after the cursor, it returns the code that belongs between them.

This is not chat

Use POST /complete for editor autocomplete — short, low-latency, no conversation. For anything conversational or agentic, use Chat Completions or Anthropic Messages.

POST/complete

Request body

prefixstringbodyrequired

Text before the cursor. An empty value short-circuits to an empty completion.

suffixstringbody

Text after the cursor. Defaults to an empty string.

languagestringbody

Language hint, e.g. typescript. Truncated to 64 characters.

file_pathstringbody

Path of the file being edited, used as additional context. Truncated to 256 characters.

Response fields

completionstring

The generated text. Always present; empty when nothing could be generated.

Request

curl -X POST https://api.misar.io/coder/complete \
  -H "Authorization: Bearer $MISARCODER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "prefix": "function slugify(input: string) {\n  return input\n    .toLowerCase()\n    ",
    "suffix": "\n}",
    "language": "typescript",
    "file_path": "src/utils/slug.ts"
  }'

200 — Completion

{ "completion": ".trim()\n    .replace(/[^a-z0-9]+/g, \"-\")\n    .replace(/^-|-$/g, \"\");" }

200 — Nothing generated

{ "completion": "" }

Failure looks like an empty completion

If the upstream model stream errors, the response is still 200 with completion: "". There is no error status for a generation failure — treat an empty string as "no suggestion", not as a signal that the request was malformed.

Behaviour

Generation is capped at 128 output tokens — this is an autocomplete, not a code generator.
Markdown code fences are stripped before the completion is returned, so you can insert the value directly into the buffer.
The response is not streamed.
The model is the deployment's configured completion model, falling back to its default model.

Limits

Limit	Value
`prefix` + `suffix` combined	32,000 characters
`language`	64 characters
`file_path`	256 characters
Output tokens	128

Status codes

Code	Meaning
`200`	Completion returned, possibly empty
`400`	`prefix + suffix too large (max 32000 chars)`
`401`	`Missing authorization` or `Invalid API key`
`422`	Body failed validation
`503`	`Router not initialized` — the deployment has no model router configured

Inline Completion

Behaviour

Limits

Status codes

On this page