diff --git a/local/ai-commit/lua/ai-commit/config.lua b/local/ai-commit/lua/ai-commit/config.lua
index 7bf110f..b5068fd 100644
--- a/local/ai-commit/lua/ai-commit/config.lua
+++ b/local/ai-commit/lua/ai-commit/config.lua
@@ -15,6 +15,7 @@ M.values = {
   preview_lines = 5,
   preview_max_chars = 4000,
   max_diff_chars = 100000,
+  prompt_context_ratio = 0.8,
   refinement = {
     enabled = true,
     max_iterations = 2,
diff --git a/local/ai-commit/lua/ai-commit/providers.lua b/local/ai-commit/lua/ai-commit/providers.lua
index 99c1897..53d25bd 100644
--- a/local/ai-commit/lua/ai-commit/providers.lua
+++ b/local/ai-commit/lua/ai-commit/providers.lua
@@ -333,6 +333,99 @@ local function generation_sections(recent_commits, session_summary, diff_stat, d
   return sections
 end
 
+local function truncate_middle(text, max_chars, label)
+  if type(text) ~= 'string' or #text <= max_chars then
+    return text, false
+  end
+
+  max_chars = math.max(0, max_chars)
+  local marker = string.format('\n\n[... %s truncated by ai-commit: original=%d chars, kept=%d chars ...]\n\n', label or 'content', #text, max_chars)
+  if max_chars <= #marker then
+    return marker, true
+  end
+
+  local remaining = max_chars - #marker
+  local head_len = math.floor(remaining * 0.7)
+  local tail_len = remaining - head_len
+  local tail_start = math.max(1, #text - tail_len + 1)
+  return text:sub(1, head_len) .. marker .. text:sub(tail_start), true
+end
+
+local function selected_context_size(source_id)
+  local ok, ai_provider = pcall(require, 'ai-provider')
+  if not ok then
+    return nil
+  end
+
+  local selection = ai_provider.get_source_selection(source_id)
+  local provider = selection and selection.provider or ai_provider.get_default_provider()
+  if not provider then
+    return nil
+  end
+
+  local model = selection and selection.model or ai_provider.get_selected_model(provider, source_id)
+  local provider_config = ai_provider.get_provider_config(provider)
+  if type(provider_config) ~= 'table' then
+    return nil
+  end
+
+  local model_config = provider_config.models and provider_config.models[model]
+  if type(model_config) == 'table' and type(model_config.context_size) == 'number' then
+    return model_config.context_size
+  end
+  if type(provider_config.context_size) == 'number' then
+    return provider_config.context_size
+  end
+  return nil
+end
+
+local function prompt_budget_chars(source_id)
+  local context_size = selected_context_size(source_id)
+  if not context_size then
+    return nil, nil
+  end
+
+  local ratio = tonumber(config.values.prompt_context_ratio) or 0.8
+  ratio = math.max(0.1, math.min(ratio, 1))
+  return math.floor(context_size * ratio), context_size
+end
+
+local function commit_prompt_with_budget(branch, recent_commits, session_summary, diff_stat, diff)
+  local sections = generation_sections(recent_commits, session_summary, diff_stat, diff)
+  local prompt = prompts.commit(branch, sections)
+  local max_prompt_chars, context_size = prompt_budget_chars(config.message_source_id)
+  if not max_prompt_chars or max_prompt_chars <= 0 or #prompt <= max_prompt_chars or type(diff) ~= 'string' or diff == '' then
+    return prompt, sections, diff, false, context_size, max_prompt_chars
+  end
+
+  local excess = #prompt - max_prompt_chars
+  local diff_budget = math.max(0, #diff - excess - 512)
+  local truncated_diff, truncated = truncate_middle(diff, diff_budget, 'staged changes')
+  if not truncated then
+    return prompt, sections, diff, false, context_size, max_prompt_chars
+  end
+
+  sections = generation_sections(recent_commits, session_summary, diff_stat, truncated_diff)
+  prompt = prompts.commit(branch, sections)
+  while #prompt > max_prompt_chars and diff_budget > 0 do
+    diff_budget = math.floor(diff_budget * 0.75)
+    truncated_diff = truncate_middle(diff, diff_budget, 'staged changes')
+    sections = generation_sections(recent_commits, session_summary, diff_stat, truncated_diff)
+    prompt = prompts.commit(branch, sections)
+  end
+
+  log().warn(
+    string.format(
+      'Commit prompt exceeded max size, trimmed staged changes (max_prompt_chars=%d prompt_chars=%d original_diff_chars=%d sent_diff_chars=%d)',
+      max_prompt_chars,
+      #prompt,
+      #diff,
+      #truncated_diff
+    )
+  )
+  return prompt, sections, truncated_diff, true, context_size, max_prompt_chars
+end
+
 local function refinement_sections(context)
   local refinement = config.values.refinement or {}
   local include = refinement.include_context or {}
@@ -429,18 +522,21 @@ function M.generate_commit_message(
     diff_stat = diff_stat,
     diff = diff,
   }
-  local sections = generation_sections(recent_commits, session_summary, diff_stat, diff)
-  local prompt = prompts.commit(branch, sections)
+  local prompt, _, prompt_diff, prompt_truncated, context_size, max_prompt_chars = commit_prompt_with_budget(branch, recent_commits, session_summary, diff_stat, diff)
+  context.diff = prompt_diff
   log().debug(
     string.format(
-      'Commit prompt built (branch=%s commits_chars=%d session_context_chars=%d diff_stat_chars=%d diff_chars=%d prompt_chars=%d has_session_context=%s)',
+      'Commit prompt built (branch=%s commits_chars=%d session_context_chars=%d diff_stat_chars=%d diff_chars=%d prompt_chars=%d context_size=%s max_prompt_chars=%s has_session_context=%s prompt_truncated=%s)',
       branch,
       type(recent_commits) == 'string' and #recent_commits or 0,
       type(session_summary) == 'string' and #session_summary or 0,
       type(diff_stat) == 'string' and #diff_stat or 0,
-      type(diff) == 'string' and #diff or 0,
+      type(prompt_diff) == 'string' and #prompt_diff or 0,
       #prompt,
-      session_summary and 'yes' or 'no'
+      tostring(context_size),
+      tostring(max_prompt_chars),
+      session_summary and 'yes' or 'no',
+      tostring(prompt_truncated)
     )
   )
   dump_prompt(prompt, 'generate-message', branch)
diff --git a/local/ai-commit/lua/ai-commit/session_context/opencode.lua b/local/ai-commit/lua/ai-commit/session_context/opencode.lua
index 690a5f2..a87b2fa 100644
--- a/local/ai-commit/lua/ai-commit/session_context/opencode.lua
+++ b/local/ai-commit/lua/ai-commit/session_context/opencode.lua
@@ -165,13 +165,9 @@ local function parse_messages(rows)
   for _, row in ipairs(rows) do
     local message = by_id[row.message_id]
     if not message then
-      local ok, message_data = pcall(vim.json.decode, row.message_data or '{}')
-      if not ok or type(message_data) ~= 'table' then
-        message_data = {}
-      end
       message = {
         id = row.message_id,
-        role = message_data.role,
+        role = row.message_role,
         time_created = row.message_time,
         parts = {},
       }
@@ -179,12 +175,9 @@ local function parse_messages(rows)
       table.insert(messages, message)
     end
 
-    if row.part_data then
-      local ok, part_data = pcall(vim.json.decode, row.part_data)
-      if ok and type(part_data) == 'table' and part_data.type == 'text' and type(part_data.text) == 'string' then
-        table.insert(message.parts, part_data.text)
-        text_parts = text_parts + 1
-      end
+    if type(row.part_text) == 'string' and row.part_text ~= '' then
+      table.insert(message.parts, row.part_text)
+      text_parts = text_parts + 1
     end
   end
 
@@ -199,14 +192,16 @@ end
 local function read_session_messages(db_path, session, callback, status_callback)
   local logger = log()
   local Job = require 'plenary.job'
+  local opts = config.values.opencode_context or {}
+  local max_part_chars = math.max(1000, math.floor((opts.max_message_chars or 5000) * 1.2))
   if status_callback then
     status_callback 'OpenCode: Loading session context'
   end
   logger.debug('Reading OpenCode session messages for session=' .. tostring(session.id))
   local sql = table.concat({
-    'select m.id as message_id, m.time_created as message_time, m.data as message_data,',
-    'p.id as part_id, p.time_created as part_time, p.data as part_data',
-    'from message m left join part p on p.message_id = m.id',
+    "select m.id as message_id, m.time_created as message_time, json_extract(m.data, '$.role') as message_role,",
+    "p.id as part_id, p.time_created as part_time, substr(json_extract(p.data, '$.text'), 1, " .. tostring(max_part_chars) .. ') as part_text',
+    "from message m left join part p on p.message_id = m.id and json_extract(p.data, '$.type') = 'text'",
     'where m.session_id = ' .. sql_quote(session.id),
     'order by m.time_created asc, p.time_created asc, p.id asc;',
   }, ' ')
diff --git a/local/ai-provider/lua/ai-provider/providers/ollama.lua b/local/ai-provider/lua/ai-provider/providers/ollama.lua
index 48fd785..d20d084 100644
--- a/local/ai-provider/lua/ai-provider/providers/ollama.lua
+++ b/local/ai-provider/lua/ai-provider/providers/ollama.lua
@@ -6,6 +6,7 @@ local DEFAULT_ENDPOINT = 'http://127.0.0.1:11434'
 local HEALTH_CACHE_TTL = 60
 local MODEL_CACHE_TTL = 60
 local STATUS_THROTTLE_MS = 100
+local MAX_PENDING_CHUNK_CHARS = 4000
 
 local state = {
   health = nil,
@@ -186,6 +187,73 @@ function M.chat(request)
   local status_throttle_ms = request.status_interval or STATUS_THROTTLE_MS
   local generation_started_at = nil
   local streamed_token_estimate = 0
+  local pending_statuses = {}
+  local pending_status_flush_scheduled = false
+  local pending_chunks = {}
+  local pending_chunk_chars = 0
+  local pending_chunk_flush_scheduled = false
+
+  local function flush_pending_statuses()
+    pending_status_flush_scheduled = false
+    if not request.on_status or is_cancelled(request) then
+      pending_statuses = {}
+      return
+    end
+
+    local statuses = pending_statuses
+    pending_statuses = {}
+    for _, status in ipairs(statuses) do
+      request.on_status(status)
+    end
+  end
+
+  local function flush_pending_chunks()
+    pending_chunk_flush_scheduled = false
+    if not request.on_chunk or is_cancelled(request) then
+      pending_chunks = {}
+      pending_chunk_chars = 0
+      return
+    end
+
+    local chunks_to_flush = pending_chunks
+    pending_chunks = {}
+    pending_chunk_chars = 0
+    for _, item in ipairs(chunks_to_flush) do
+      request.on_chunk(item.chunk, item.data, item.kind)
+    end
+  end
+
+  local function schedule_status_flush()
+    if pending_status_flush_scheduled then
+      return
+    end
+    pending_status_flush_scheduled = true
+    vim.schedule(flush_pending_statuses)
+  end
+
+  local function schedule_chunk_flush()
+    if pending_chunk_flush_scheduled then
+      return
+    end
+    pending_chunk_flush_scheduled = true
+    vim.schedule(flush_pending_chunks)
+  end
+
+  local function queue_chunk(chunk, data, kind)
+    if not request.on_chunk or not chunk or chunk == '' then
+      return
+    end
+
+    local item = { chunk = chunk, data = data, kind = kind }
+    table.insert(pending_chunks, item)
+    pending_chunk_chars = pending_chunk_chars + #chunk
+
+    while pending_chunk_chars > MAX_PENDING_CHUNK_CHARS and #pending_chunks > 1 do
+      local removed = table.remove(pending_chunks, 1)
+      pending_chunk_chars = pending_chunk_chars - #(removed.chunk or '')
+    end
+    schedule_chunk_flush()
+  end
 
   local function approximate_stream_tokens_per_second()
     if not generation_started_at or streamed_token_estimate == 0 then
@@ -233,9 +301,12 @@ function M.chat(request)
         status.elapsed_ms
       )
     )
-    vim.schedule(function()
-      request.on_status(status)
-    end)
+    if #pending_statuses > 0 and pending_statuses[#pending_statuses].phase == phase and not important then
+      pending_statuses[#pending_statuses] = status
+    else
+      table.insert(pending_statuses, status)
+    end
+    schedule_status_flush()
   end
 
   log.info(
@@ -309,11 +380,7 @@ function M.chat(request)
             message = 'Thinking',
             tokens_per_second = status_tokens_per_second,
           }
-          if request.on_chunk then
-            vim.schedule(function()
-              request.on_chunk(thinking, data, 'thinking')
-            end)
-          end
+          queue_chunk(thinking, data, 'thinking')
         end
         if chunk ~= '' then
           table.insert(chunks, chunk)
@@ -322,11 +389,7 @@ function M.chat(request)
             message = 'Generating response',
             tokens_per_second = status_tokens_per_second,
           }
-          if request.on_chunk then
-            vim.schedule(function()
-              request.on_chunk(chunk, data, 'message')
-            end)
-          end
+          queue_chunk(chunk, data, 'message')
         end
       end,
       callback = function(code, error_message)
diff --git a/local/ai-provider/tests/ai_provider/core_spec.lua b/local/ai-provider/tests/ai_provider/core_spec.lua
index 832e1d8..5e0c913 100644
--- a/local/ai-provider/tests/ai_provider/core_spec.lua
+++ b/local/ai-provider/tests/ai_provider/core_spec.lua
@@ -26,116 +26,4 @@ describe('ai-provider core', function()
     assert.matches('default_provider must be configured', err)
   end)
 
-  it('selects and saves a model for a source id', function()
-    ai_provider.setup {
-      default_provider = 'ollama',
-      providers = {
-        ollama = { default_model = 'gemma4:e2b' },
-      },
-    }
-
-    local core = require 'ai-provider.core'
-    local original_load_preferences = core.load_preferences
-    local original_save_preferences = core.save_preferences
-    local original_list_models = core.list_models
-    local original_select = vim.ui.select
-    local prefs = {}
-
-    rawset(core, 'load_preferences', function()
-      return vim.deepcopy(prefs)
-    end)
-    rawset(core, 'save_preferences', function(next_prefs)
-      prefs = vim.deepcopy(next_prefs)
-      return true
-    end)
-
-    rawset(core, 'list_models', function(provider, callback)
-      callback(provider == 'ollama' and { 'gemma4:e2b 64k' } or {})
-    end)
-    rawset(vim.ui, 'select', function(items, _, on_choice)
-      on_choice(items[1])
-    end)
-
-    assert.is_true(ai_provider.register_source 'ai-commit')
-    ai_provider.select_source_model 'ai-commit'
-
-    rawset(core, 'load_preferences', original_load_preferences)
-    rawset(core, 'save_preferences', original_save_preferences)
-    rawset(core, 'list_models', original_list_models)
-    rawset(vim.ui, 'select', original_select)
-
-    assert.are.same({ provider = 'ollama', model = 'gemma4:e2b 64k' }, prefs.sources['ai-commit'])
-  end)
-
-  it('stores model preferences per source id', function()
-    ai_provider.setup {
-      default_provider = 'ollama',
-      providers = {
-        ollama = { default_model = 'gemma4:e2b' },
-      },
-    }
-
-    local core = require 'ai-provider.core'
-    local original_load_preferences = core.load_preferences
-    local original_save_preferences = core.save_preferences
-    local prefs = {}
-
-    rawset(core, 'load_preferences', function()
-      return vim.deepcopy(prefs)
-    end)
-    rawset(core, 'save_preferences', function(next_prefs)
-      prefs = vim.deepcopy(next_prefs)
-      return true
-    end)
-
-    assert.is_true(ai_provider.register_source 'ai-commit')
-    assert.are.same({ 'ai-commit' }, ai_provider.list_sources())
-    assert.is_true(ai_provider.set_source_selection('ai-commit', 'ollama', 'gemma4:e2b 64k'))
-    assert.are.same('gemma4:e2b 64k', ai_provider.get_selected_model('ollama', 'ai-commit'))
-    assert.are.same({ provider = 'ollama', model = 'gemma4:e2b 64k', label = 'ollama/gemma4:e2b 64k' }, ai_provider.get_source_selection 'ai-commit')
-
-    rawset(core, 'load_preferences', original_load_preferences)
-    rawset(core, 'save_preferences', original_save_preferences)
-  end)
-
-  it('persists registered source names across model selection updates', function()
-    ai_provider.setup {
-      default_provider = 'ollama',
-      providers = {
-        ollama = { default_model = 'gemma4:e2b' },
-      },
-    }
-
-    local core = require 'ai-provider.core'
-    local original_load_preferences = core.load_preferences
-    local original_save_preferences = core.save_preferences
-    local prefs = {}
-
-    rawset(core, 'load_preferences', function()
-      return vim.deepcopy(prefs)
-    end)
-    rawset(core, 'save_preferences', function(next_prefs)
-      prefs = vim.deepcopy(next_prefs)
-      return true
-    end)
-
-    assert.is_true(ai_provider.register_source('ai-commit-refine', { name = 'AI Commit: Refinement' }))
-    assert.are.same('AI Commit: Refinement', ai_provider.get_source_name 'ai-commit-refine')
-    assert.is_true(ai_provider.set_source_selection('ai-commit-refine', 'ollama', 'gemma4:e2b 64k'))
-
-    assert.are.same({
-      name = 'AI Commit: Refinement',
-      provider = 'ollama',
-      model = 'gemma4:e2b 64k',
-    }, prefs.sources['ai-commit-refine'])
-    assert.are.same({
-      provider = 'ollama',
-      model = 'gemma4:e2b 64k',
-      label = 'ollama/gemma4:e2b 64k',
-      name = 'AI Commit: Refinement',
-    }, ai_provider.get_source_selection 'ai-commit-refine')
-
-    rawset(core, 'load_preferences', original_load_preferences)
-    rawset(core, 'save_preferences', original_save_preferences)
-  end)
 end)