omni-dev 0.44.0

AI-powered git commit rewriter, PR generator, and MCP server for Jira, Confluence, Datadog, Gmail, and Drive.
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103

# Effort capabilities checked 2026-09-23; criteria are uncalibrated routing hypotheses.
# The `openai` model ladder for `omni-dev ai jev route` (#1820):
# terra (gpt-5.6-terra) / sol (gpt-5.6-sol) / astra (gpt-6-astra).
#
# Tiers are listed from least to most capable; `route` ranks them by position.
# The descriptions are the anthropic ladder's, rung for rung, byte-identical:
# that text is the only text validated against jev-1.13.0 (#1779), and
# rewording shifts every answer. Only the tier names differ. This ladder has
# been compared in docs/evaluations/jev-effort-1888; downstream success
# remains uncalibrated, so treat its answers as a starting point.
tiers:
  - name: terra
    models:
      - name: gpt-5.6-terra
        verified: "2026-09-23"
        sources: [https://developers.openai.com/api/docs/models/gpt-5.6-terra]
        effort:
          kind: configurable
          control: reasoning.effort
          supported_levels: [none, low, medium, high, xhigh, max]
          levels:
            - name: none
              description: >-
                No explicit reasoning is needed: direct mechanical edits or exact application of a supplied checklist with no unresolved interactions.
            - name: low
              description: >-
                The stage is short, local and familiar, with explicit requirements and an obvious solution or review checklist.
            - name: medium
              description: >-
                The stage needs several reasoning steps and checking ordinary interactions, with a clear approach and bounded scope.
            - name: high
              description: >-
                The stage needs sustained reasoning about subtle behavior, multiple components, non-obvious failure modes or competing approaches.
            - name: xhigh
              description: >-
                The stage needs extended exploration, long-horizon implementation or review, or repeated testing of interacting hypotheses beyond high effort.
            - name: max
              description: >-
                The stage approaches this model's capability limit and requires its deepest supported analysis; lower effort would leave major unresolved risks.
    description: >-
      Reliable at executing a clear specification and following patterns that already exist in the
      codebase. Tends to miss non-obvious interactions between parts and does not question the
      specification.
  - name: sol
    models:
      - name: gpt-5.6-sol
        verified: "2026-09-23"
        sources: [https://developers.openai.com/api/docs/models/gpt-5.6-sol]
        effort:
          kind: configurable
          control: reasoning.effort
          supported_levels: [none, low, medium, high, xhigh, max]
          levels:
            - name: none
              description: >-
                No explicit reasoning is needed: direct mechanical edits or exact application of a supplied checklist with no unresolved interactions.
            - name: low
              description: >-
                The stage is short, local and familiar, with explicit requirements and an obvious solution or review checklist.
            - name: medium
              description: >-
                The stage needs several reasoning steps and checking ordinary interactions, with a clear approach and bounded scope.
            - name: high
              description: >-
                The stage needs sustained reasoning about subtle behavior, multiple components, non-obvious failure modes or competing approaches.
            - name: xhigh
              description: >-
                The stage needs extended exploration, long-horizon implementation or review, or repeated testing of interacting hypotheses beyond high effort.
            - name: max
              description: >-
                The stage approaches this model's capability limit and requires its deepest supported analysis; lower effort would leave major unresolved risks.
    description: >-
      Reliable at reasoning across many files, subtle library or platform semantics, concurrency and
      ordering, once the overall direction is set. Can under-explore a genuinely open design space.
  - name: astra
    models:
      - name: gpt-6-astra
        verified: "2026-09-23"
        sources: [https://developers.openai.com/api/docs/models/gpt-6-astra]
        effort:
          kind: configurable
          control: reasoning.effort
          supported_levels: [low, medium, high, xhigh, max]
          levels:
            - name: low
              description: >-
                The stage is short, local and familiar, with explicit requirements and an obvious solution or review checklist.
            - name: medium
              description: >-
                The stage needs several reasoning steps and checking ordinary interactions, with a clear approach and bounded scope.
            - name: high
              description: >-
                The stage needs sustained reasoning about subtle behavior, multiple components, non-obvious failure modes or competing approaches.
            - name: xhigh
              description: >-
                The stage needs extended exploration, long-horizon implementation or review, or repeated testing of interacting hypotheses beyond high effort.
            - name: max
              description: >-
                The stage approaches this model's capability limit and requires its deepest supported analysis; lower effort would leave major unresolved risks.
    description: >-
      Strongest at open-ended design and research: choosing between architectures, working in
      unfamiliar territory with no precedent in the codebase, anticipating failure modes, and
      security-critical judgement.