ollama_summarize_deep

Summarize (deep). Digest of long input with optional focus. Pass EXACTLY ONE of: text (raw content in hand), source_path (single file — server reads + chunks, Claude never pre-reads), or source_paths[] (multiple files). The path-based shapes save Claude context — the whole point of delegating summarization. Carries source_preview for fabrication spot-checks. FRAME CONTRACT (optional): pass frame (the question / section purpose this digest is FOR — distinct from focus, which is emphasis WITHIN an in-frame source). Sources that don’t address the frame are dropped from the digest and listed under unaddressed_sources; if NO source addresses it, summary=‘’ and frame_addressed: false. Omitting frame preserves the legacy result shape.

Parameters

Parameter	Type	Required	Default	Description
text	string	no	—	Raw text to digest. Use this when you already have the content in hand.
source_path	string	no	—	A single file path to read + chunk server-side. Shortcut for source_paths: [one] — use whichever reads more naturally. Closes adoption SEAM #2 (large single-file summaries no longer force Claude to pre-read the file).
source_paths	any	no	—	File paths to read + chunk server-side. Use this instead of text to save Claude context — the tool reads the files, Claude never sees the raw content.
focus	string	no	—	Optional aspect to emphasize (e.g. ‘combat doctrine’, ‘auth flow’).
frame	string	no	—	The question / section purpose / topic this digest is FOR — distinct from focus, which is emphasis within an in-frame source. When supplied, sources that don’t address the frame are dropped from the digest and listed under unaddressed_sources instead. If NO source addresses the frame, summary is empty.
max_words	integer	no	—	Target digest length in words (default 250).
per_file_max_chars	integer	no	—	Chars to read per file when source_paths is used (default 40k).
model	string	no	—	Optional per-call model override. When provided, overrides the tool’s tier-resolved model for this call. The tier’s timeout (TIER_TIMEOUT_MS) still applies. On timeout, fallback uses the tier-resolved model, NOT the override. Use for receipt-backed orchestration that requires explicit model identity (e.g., research-os reviewer profiles).

Full input schema

The JSON Schema below is generated from the same zod schema the server validates against at the wire — it cannot drift from runtime behavior.

{
  "$schema": "https://json-schema.org/draft/2020-12/schema",
  "type": "object",
  "properties": {
    "text": {
      "description": "Raw text to digest. Use this when you already have the content in hand.",
      "type": "string",
      "minLength": 1
    },
    "source_path": {
      "description": "A single file path to read + chunk server-side. Shortcut for `source_paths: [one]` — use whichever reads more naturally. Closes adoption SEAM #2 (large single-file summaries no longer force Claude to pre-read the file).",
      "type": "string",
      "minLength": 1
    },
    "source_paths": {
      "description": "File paths to read + chunk server-side. Use this instead of `text` to save Claude context — the tool reads the files, Claude never sees the raw content."
    },
    "focus": {
      "description": "Optional aspect to emphasize (e.g. 'combat doctrine', 'auth flow').",
      "type": "string"
    },
    "frame": {
      "description": "The question / section purpose / topic this digest is FOR — distinct from `focus`, which is emphasis within an in-frame source. When supplied, sources that don't address the frame are dropped from the digest and listed under `unaddressed_sources` instead. If NO source addresses the frame, summary is empty.",
      "type": "string"
    },
    "max_words": {
      "description": "Target digest length in words (default 250).",
      "type": "integer",
      "minimum": 20,
      "maximum": 1500
    },
    "per_file_max_chars": {
      "description": "Chars to read per file when source_paths is used (default 40k).",
      "type": "integer",
      "minimum": 1000,
      "maximum": 200000
    },
    "model": {
      "description": "Optional per-call model override. When provided, overrides the tool's tier-resolved model for this call. The tier's timeout (TIER_TIMEOUT_MS) still applies. On timeout, fallback uses the tier-resolved model, NOT the override. Use for receipt-backed orchestration that requires explicit model identity (e.g., research-os reviewer profiles).",
      "type": "string",
      "minLength": 1
    }
  }
}

Envelope

Every tool returns the standard envelope — tier_used, model, tokens_in/tokens_out, elapsed_ms, backend provenance, warnings. See Envelope & tiers.

Source

Schema + handler: src/tools/summarizeDeep.ts · registration: src/index.ts

---

Back to the Tool Reference.