Your Agents Need a Save Button - Hamza Tahir, ZenML

What was stored

cues

171

whisperx 171

chunks

31

from 171 cues

keyframes

89

kept of 135 captured

frames with text

89

5,137 lines read

chapters

0

from the source metadata

keyframe bytes

20.3 MB

word timings on 171 cues

Provenance

Frames, and what the machine read

• 

  0:04 #0 done5 line(s)

  shot 0·sharpness 1124.7

  1. oKitaru0.89
  2. What if you could ask "what if?"0.97
  3. about a run that already happened?1.00
  4. Change one thing. Replay it. See what would have happened — no0.98
  5. re-running in production.1.00
• 

  0:18 #1 done3 line(s)

  shot 1·sharpness 628.2

  1. oKitaru0.88
  2. Your agents need a save button.1.00
  3. Hamza Tahir· Co-founder of ZenML·Building Kitaru1.00
• 

  0:30 #2 done9 line(s)

  shot 2·sharpness 1290.8

  1. CHAPTER 1 · THE EXECUTION IS GONE0.96
  2. We've had the save button since the early 1980s.0.98
  3. So why don't our agents have them?1.00
  4. FILE·19841.00
  5. 9eS0.80
  6. AGENT·20260.99
  7. + no save0.95
  8. State persists. Close it, reopen it, keep working.1.00
  9. State lives in the process. When it exits, gone.1.00
• 

  0:44 #3 done19 line(s)

  shot 3·sharpness 1476.4

  1. CHAPTER 10.99
  2. A trace is a description, held by a system that never0.99
  3. ran your code.0.98
  4. Your execution ran in your process. The telemetry went somewhere else — Braintrust, Langfuse, LangSmith.1.00
  5. Nothing connects the two.1.00
  6. OBSERVABILITY BACKEND - off to the side0.98
  7. run · draft-flow · completed ENDED 14:09:520.96
  8. YOUR PROCESS0.99
  9. The execution1.00
  10. spans·events1.00
  11. research1.00
  12. tool calls run here0.97
  13. ONE-WAY1.00
  14. write_draft1.00
  15. READ-ONLY1.00
  16. real state lives here1.00
  17. publish1.00
  18. then it's gone0.97
  19. done1.00
• 

  1:03 #4 skipped

  shot 4·duplicate of #3

• 

  1:31 #5 done17 line(s)

  shot 5·sharpness 1037.9

  1. CHAPTER 21.00
  2. Agents have many dials - simply choose0.97
  3. your new configuration and replay.1.00
  4. one replay call1.00
  5. {}0.99
  6. X0.99
  7. Swap the model0.99
  8. Mock a tool1.00
  9. Fail a call1.00
  10. Degrade1.00
  11. Drop a tool0.98
  12. retriever1.00
  13. overrides={model}1.00
  14. overrides={tool}1.00
  15. raise on step0.97
  16. swap the index1.00
  17. remove from set0.98
• 

  2:26 #6 done12 line(s)

  shot 6·sharpness 1258.0

  1. CHAPTER 21.00
  2. Add a durable runtime layer.1.00
  3. Augment your traces with true checkpoints .0.99
  4. Claude1.00
  5. LangGraph1.00
  6. python1.00
  7. Pydantic1.00
  8. A Google Antigravity0.98
  9. OpenAI Agents SDK0.99
  10. Agent SDK1.00
  11. Durable runtime1.00
  12. checkpoints every call underneath0.99
• 

  2:32 #7 done14 line(s)

  shot 7·sharpness 1099.2

  1. CHAPTER 3· YOUR TEST SET WROTE ITSELF0.99
  2. Your production runs are the ultimate test set.1.00
  3. Synthetic cases imagine what users might do. Real runs are what they did — real inputs, real tool responses, real0.99
  4. edge cases, including the ones that already went wrong.1.00
  5. SYNTHETIC1.00
  6. PRODUCTION1.00
  7. imagined - what users might do0.98
  8. real runs - what they actually did0.99
  9. Customer asks refund0.98
  10. Refund after a chargeback dispute1.00
  11. Check my order status0.99
  12. Order status - in Spanish, mid-chat0.99
  13. Reset my password - request0.99
  14. Refund request, tool timed out0.98
• 

  3:01 #8 skipped

  shot 8·duplicate of #7

• 

  3:33 #9 done17 line(s)

  shot 9·sharpness 1171.5

  1. CHAPTER 3 · THE LOOP0.97
  2. Close the loop and turn past runs into1.00
  3. full-blown evals.1.00
  4. Orepeat0.98
  5. 21.00
  6. 31.00
  7. 41.00
  8. Select a cohort0.98
  9. Replay one1.00
  10. Diff1.00
  11. Decide1.00
  12. change1.00
  13. the runs that matter0.98
  14. same call, one keyword1.00
  15. cost · decisions0.94
  16. divergence1.00
  17. ship · route · hold0.90
• 

  4:00 #10 done12 line(s)

  shot 10·sharpness 1044.2

  1. CHAPTER 31.00
  2. Four steps. The methodology is the skill.0.99
  3. Checkpoint1.00
  4. the expensive, the escalated, the failed.0.99
  5. Replay1.00
  6. everything before your change, handed back from cache.0.99
  7. 030.98
  8. Diff0.98
  9. cost, changed decisions, where it first diverged.0.99
  10. 040.99
  11. Decide1.00
  12. ship what matches, route what's marginal, hold what drifts.1.00
• 

  4:06 #11 skipped

  shot 11·duplicate of #9

• 

  4:15 #12 skipped

  shot 12·duplicate of #10

• 

  4:21 #13 done11 line(s)

  shot 13·sharpness 1335.0

  1. CHAPTER 4 · THIS WORKS AT SCALE0.98
  2. Don't believe me? Doordash and others1.00
  3. are doing it too.1.00
  4. DOORDASH1.00
  5. Blog0.95
  6. Inside DoorDash's one-click simulation and evaluation1.00
  7. platform for support chatbots0.99
  8. Joune I 20Chanan GoAas ambs0.66
  9. https://careersatdoordash.com/blog/doordashs-one1.00
  10. -click-simulation-and-evaluation-platform-for-sup1.00
  11. port-chatbots/1.00
• 

  4:45 #14 done16 line(s)

  shot 14·sharpness 746.0

  1. CHAPTER 4 · DOORDASH0.98
  2. 3021.00
  3. 21.00
  4. -901.00
  5. sims1.00
  6. pts1.00
  7. %0.97
  8. in 5 minutes0.93
  9. within target1.00
  10. hallucinations1.00
  11. where 1% of live traffic0.99
  12. simulated escalation rate1.00
  13. once the eval flywheel1.00
  14. took 7 hours0.97
  15. vs production1.00
  16. was in place1.00
• 

  4:59 #15 done4 line(s)

  shot 15·sharpness 578.5

  1. DEMO1.00
  2. One run → ten → ask the0.96
  3. agent.1.00
  4. A support agent in production. Every model and tool call checkpointed.0.99
• 

  5:11 #16 done76 line(s)

  shot 16·sharpness 3154.1

  1. Chrome1.00
  2. File1.00
  3. Edit1.00
  4. View1.00
  5. History1.00
  6. Bookmarks1.00
  7. Profiles1.00
  8. Tab1.00
  9. Window1.00
  10. Help1.00
  11. 70.77
  12. V0.99
  13. A1.00
  14. 4分0.55
  15. Q0.96
  16. Thu 25. Jun 11:090.97
  17. Executions - Kitaru0.97
  18. 口0.74
  19. Your Agents Need a Save But ×0.98
  20. zenml-io/kitaru: Open-source ×0.99
  21. +0.54
  22. ←1.00
  23. →0.97
  24. C0.69
  25. github.com/zenml-io/kitaru1.00
  26. ☆1.00
  27. le0.81
  28. 米0.69
  29. Work1.00
  30. New Chrome available0.98
  31. GETTING_STARTED.md1.00
  32. Migrate Kitaru docs to GitBook + standalone SDK referenc...0.99
  33. 3 weeks ago1.00
  34. Justfile1.00
  35. Add sandbox stack and adapter command support (#423)1.00
  36. last week1.00
  37. Languages1.00
  38. LICENSE1.00
  39. Initial commit1.00
  40. 3 months ago1.00
  41. Python 98.2%0.96
  42. Shell 1.5%1.00
  43. Other 0.3%1.00
  44. README.md1.00
  45. Point Agent Harness Platform docs at the ZenML Agents g..0.99
  46. 2 days ago1.00
  47. SECURITY.md1.00
  48. Add security policy for vulnerability reporting0.99
  49. 3 months ago1.00
  50. lychee.toml1.00
  51. Point Agent Harness Platform docs at the ZenML Agents g...0.98
  52. 2 days ago1.00
  53. pyproject.toml1.00
  54. Release 0.18.01.00
  55. 1 hour ago1.00
  56. uv.lock1.00
  57. Release 0.18.01.00
  58. 1 hour ago1.00
  59. wrangler.redirect.toml1.00
  60. Migrate Kitaru docs to GitBook + standalone SDK referenc...0.98
  61. 3 weeks ago1.00
  62. wrangler.toml1.00
  63. Migrate Kitaru docs to GitBook + standalone SDK referenc...1.00
  64. 3 weeks ago1.00
  65. README1.00
  66. 8 Contributing0.95
  67. 8 Apache-2.0 license0.97
  68. Security1.00
  69. 00.95
  70. 三0.96
  71. Kitaru1.00
  72. The runtime layer underneath your agent stack.0.99
  73. Kitaru (来る, "to arrive") is a self-hosted, framework-agnostic runtime for autonomous agents — underneath the0.99
  74. harness your team already picked. You keep your agent SDK, your prompts, your tools, your model. Kitaru adds0.99
  75. durable execution: checkpoints. replav. resume. wait(). versioned deplovments. and isolated runtimes. runnina on0.99
  76. $10.93
• 

  5:17 #17 done79 line(s)

  shot 17·sharpness 3154.8

  1. Chrome1.00
  2. File1.00
  3. Edit1.00
  4. View1.00
  5. History1.00
  6. Bookmarks1.00
  7. Profiles1.00
  8. Tab1.00
  9. Window1.00
  10. Help1.00
  11. V1.00
  12. A1.00
  13. G0.77
  14. Q0.97
  15. Thu 25. Jun 11:100.98
  16. Executions - Kitaru0.98
  17. x0.63
  18. 口0.86
  19. Your Agents Need a Save But ×0.97
  20. zenml-io/kitaru: Open-source ×0.96
  21. +0.57
  22. ←1.00
  23. →0.98
  24. C0.88
  25. github.com/zenml-io/kitaru1.00
  26. ☆1.00
  27. le0.81
  28. 米0.72
  29. Work1.00
  30. New Chrome available0.98
  31. pyproject.toml1.00
  32. Release 0.18.01.00
  33. 1 hour ago1.00
  34. uv.lock1.00
  35. Release 0.18.01.00
  36. 1 hour ago0.99
  37. wrangler.redirect.toml1.00
  38. Migrate Kitaru docs to GitBook + standalone SDK referenc...1.00
  39. 3 weeks ago1.00
  40. wrangler.toml1.00
  41. Migrate Kitaru docs to GitBook + standalone SDK referenc...1.00
  42. 3 weeks ago1.00
  43. README1.00
  44. 83 Contributing0.95
  45. Apache-2.0 license0.99
  46. Security0.97
  47. 三0.92
  48. 00.97
  49. Kitaru1.00
  50. The runtime layer underneath your agent stack.0.99
  51. Kitaru (来る, "to arrive") is a self-hosted, framework-agnostic runtime for autonomous agents — underneath the0.99
  52. harness your team already picked. You keep your agent SDK, your prompts, your tools, your model. Kitaru adds0.99
  53. durable execution: checkpoints, replay, resume, wait(), versioned deployments, and isolated runtimes, running on0.99
  54. your own infrastructure.1.00
  55. pypi1.00
  56. v0.17.11.00
  57. python 3.11 | 3.12 | 3.130.98
  58. license1.00
  59. Apache-2.01.00
  60. Docs·Quick Start·Examples·Getting Started Guide·1.00
  61. Roadmap0.96
  62. Community1.00
  63. kitaru0.84
  64. content_pipeline ACTIVE0.96
  65. Flows1.00
  66. Overview1.00
  67. Execution0.99
  68. Logs Cost Diff1.00
  69. PIPELI content_pipeline0.88
  70. 1260.89
  71. FATLIRES 230.62
  72. COT $0.0480.85
  73. content_pipeline1.00
  74. 1261.00
  75. ICCESS RATE0.77
  76. 73%1.00
  77. $0.0400.90
  78. 10.2s1.00
  79. $10.93
• 

  5:30 #18 done181 line(s)

  shot 18·sharpness 1570.2

  1. Chrome1.00
  2. File1.00
  3. Edit1.00
  4. View1.00
  5. History1.00
  6. Bookmarks1.00
  7. Profiles1.00
  8. Tab1.00
  9. Window1.00
  10. Help1.00
  11. V0.99
  12. A1.00
  13. G0.74
  14. Q0.98
  15. Thu 25. Jun 11:100.99
  16. Executions - Kitaru0.98
  17. x0.64
  18. 口0.85
  19. Your Agents Need a Save But1.00
  20. zenml-io/kitaru: Open-source1.00
  21. ×0.68
  22. +0.55
  23. ←1.00
  24. →0.97
  25. C0.89
  26. github.com/zenml-io/kitaru1.00
  27. ☆1.00
  28. le0.81
  29. 米0.50
  30. Work1.00
  31. New Chrome available1.00
  32. 日0.59
  33. README1.00
  34. 83 Contributing0.94
  35. 8 Apache-2.0 license0.98
  36. Security1.00
  37. 三0.85
  38. 00.98
  39. Docs1.00
  40. Quick Start·0.96
  41. Examples1.00
  42. Getting Started Guide0.99
  43. Roadmap1.00
  44. Community1.00
  45. kitaru0.82
  46. Flows0.99
  47. content_pipeline0.99
  48. ACTIVE1.00
  49. Flows1.00
  50. Stacks0.92
  51. Overview1.00
  52. Execution1.00
  53. Logs1.00
  54. Cost0.99
  55. Diff0.98
  56. PIPELINE content_pipeline0.91
  57. : 260.77
  58. PATLURES 230.74
  59. ANE CT $0.0480.82
  60. content_pipeline1.00
  61. 1261.00
  62. 73%1.00
  63. $0.0400.89
  64. 10.2s0.99
  65. Completed0.99
  66. Faled0.92
  67. Waiting0.95
  68. Sleeping0.99
  69. Cancelled1.00
  70. at0.73
  71. at0.94
  72. Columns1.00
  73. EXECUTION0.95
  74. starus0.75
  75. DURATION0.99
  76. C010.54
  77. TORENI0.78
  78. CONFIO0.95
  79. DATE0.98
  80. TRIGOER0.83
  81. #1261.00
  82. COMPLETED1.00
  83. 12.2s0.96
  84. $0.0550.98
  85. 3.480.97
  86. prod0.89
  87. Mar 5, 2028 110.88
  88. schedule0.89
  89. #1250.92
  90. COMPLETED1.00
  91. 10.1s0.88
  92. 10.0370.91
  93. 5.080.83
  94. dv0.60
  95. Mar 5, 2020 17:270.93
  96. #1240.99
  97. COMPLETED1.00
  98. 18.7s0.83
  99. 80.0370.91
  100. 5.580.87
  101. prod0.98
  102. Mar 5, 2026 13:500.87
  103. #1230.89
  104. COMPLETED1.00
  105. 9.1s0.91
  106. 80.0580.93
  107. 4.180.96
  108. dev0.97
  109. Mar 5. 2026 12:270.91
  110. #1220.96
  111. COMPLETED0.99
  112. 15.6s0.87
  113. $0.0600.98
  114. $.580.75
  115. pred0.93
  116. Mer 5, 2026 13:480.87
  117. #1211.00
  118. COMPLETED1.00
  119. 10.1s0.93
  120. $0.0440.97
  121. 4.780.90
  122. prod0.98
  123. Mar 4, 2028 11030.93
  124. #1200.97
  125. COMPLETED1.00
  126. 10.8s0.95
  127. $0.0390.93
  128. 2.780.97
  129. dev0.78
  130. Mar 4, 2026 17:320.92
  131. #1190.99
  132. COMPLETED1.00
  133. 13.5s0.83
  134. $0.0390.99
  135. 4.480.78
  136. prod0.99
  137. Mar 4, 2026 17:110.94
  138. #1181.00
  139. COMPLETED1.00
  140. 11.ts0.93
  141. $0.0390.98
  142. 5.180.85
  143. dev0.83
  144. Mar 4, 2026 10:050.94
  145. #1170.99
  146. COMPLETED1.00
  147. 11.1s0.98
  148. $0.0410.95
  149. 4.580.88
  150. prod0.98
  151. Mar 3, 2026 13:080.92
  152. #1160.99
  153. FAILED0.97
  154. 1.t0.86
  155. $0.0140.95
  156. 1.780.88
  157. prod0.99
  158. Mer 2. 2020 12 280.84
  159. #1151.00
  160. CANCELLED1.00
  161. 2.0s0.96
  162. 50.0160.90
  163. 1.6K0.85
  164. prod0.98
  165. Mar 3, 2026 15 330.85
  166. #1140.84
  167. COMPLETED0.99
  168. 10.9s0.77
  169. 80.0430.91
  170. 4.180.99
  171. dev0.87
  172. Mar 3, 2026 13 390.88
  173. #1130.94
  174. COMPLETED0.98
  175. 1.500.74
  176. $0.0540.95
  177. 3.080.70
  178. dev0.61
  179. Mar 2, 2026 12 560.90
  180. Where Kitaru fits1.00
  181. $10.94
• 

  5:34 #19 done107 line(s)

  shot 19·sharpness 1805.8

  1. Chrome1.00
  2. File1.00
  3. Edit1.00
  4. View1.00
  5. History1.00
  6. Bookmarks0.99
  7. Profiles1.00
  8. Tab1.00
  9. Window1.00
  10. Help1.00
  11. 口0.50
  12. 70.86
  13. V0.99
  14. A1.00
  15. G0.75
  16. Q0.98
  17. Thu 25. Jun 11:100.98
  18. Executions - Kitaru1.00
  19. ×0.87
  20. 口0.89
  21. Your Agents Need a Save But ×0.98
  22. zenml-io/kitaru: Open-source ×0.98
  23. ←1.00
  24. →0.97
  25. C0.89
  26. preview.demo.kitaru.zenml.io/flows/cc12f224-3790-47c8-af9b-7418cf99bda3/v/local/ex...1.00
  27. ☆1.00
  28. le0.85
  29. 米0.60
  30. Work1.00
  31. New Chrome available1.00
  32. Flows1.00
  33. support_copilot_flow0.99
  34. local0.98
  35. Flows1.00
  36. Executions1.00
  37. AD1.00
  38. support_copilot_flow· not deployed0.98
  39. Executions1.00
  40. Invoke1.00
  41. local0.92
  42. All versions1.00
  43. Status1.00
  44. Stack1.00
  45. Range1.00
  46. Q Search #num or flow...1.00
  47. G0.96
  48. Refresh1.00
  49. Execution1.00
  50. Status1.00
  51. Duration1.00
  52. Cost1.00
  53. ID1.00
  54. Author1.00
  55. Created ↓0.96
  56. #00711.00
  57. Completed1.00
  58. 1m 9s1.00
  59. $0.001513251.00
  60. 9e135198-e12e-4f32-9f76-bd4cdb0ed0371.00
  61. AD1.00
  62. admin1.00
  63. 25/06/2026,10:49:351.00
  64. …0.60
  65. #00701.00
  66. Completed1.00
  67. 47s1.00
  68. $0.001387251.00
  69. 7f3e3411-f73c-4c92-bd0d-daa9c70443461.00
  70. AD1.00
  71. admin1.00
  72. 25/06/2026,10:39:541.00
  73. ...0.87
  74. #00691.00
  75. Completed1.00
  76. 37s1.00
  77. $0.000986151.00
  78. 8e07850d-d954-4c98-8541-359cbfdf1e3b1.00
  79. AD1.00
  80. admin1.00
  81. 25/06/2026,10:25:591.00
  82. ...0.63
  83. #00681.00
  84. Failed1.00
  85. 6s1.00
  86. ea434beb-2755-474c-a6f6-185a47932b511.00
  87. AD1.00
  88. admin1.00
  89. 25/06/2026,10:25:481.00
  90. ...0.75
  91. #00671.00
  92. Completed1.00
  93. 35s1.00
  94. $0.000717351.00
  95. 1a48977e-0d46-4095-b9cb-b50a33abe41d1.00
  96. AD1.00
  97. admin1.00
  98. 25/06/2026.10:25:040.99
  99. #00661.00
  100. Completed1.00
  101. 36s1.00
  102. $0.000906551.00
  103. 37689f60-1b57-4a03-a163-2d41bc9b26a71.00
  104. AD1.00
  105. admin1.00
  106. 251.00
  107. $10.94
• 

  5:35 #20 skipped

  shot 20·duplicate of #19

• 

  5:43 #21 skipped

  shot 21·duplicate of #19

• 

  5:48 #22 done18 line(s)

  shot 22·sharpness 1058.1

  1. ←1.00
  2. →0.99
  3. C0.94
  4. preview.demo.kitaru.zenml.io/flows/cc12f224-3790-47c8-af9b-7418cf99bda3/v/local/ex...1.00
  5. ☆1.00
  6. lo0.93
  7. 米0.86
  8. O0.97
  9. Flows1.00
  10. support_copilot_flow1.00
  11. local00.81
  12. #00711.00
  13. Execution1.00
  14. Logs1.00
  15. EXECUTIONS1.00
  16. List1.00
  17. Timeline1.00
  18. Filter steps...1.00
• 

  5:51 #23 done

  shot 23·sharpness 1095.3

The page's on-screen-text budget of 600 lines is spent, so the last cards in this grid list fewer lines than they hold. Narrow the page with ?frames= to read them.

Transcript

171 cues· 2,813 words· 14,686 chars