DeepSeek V4 thinking mode defaults to enabled on both deepseek-v4-flash
and deepseek-v4-pro; non-thinking behavior requires an explicit toggle.
Per official docs, in thinking mode temperature/top_p/presence_penalty/
frequency_penalty are ignored, so the existing Flash role sampling knobs
would have been silently no-ops without the toggle.
Add extra_body.thinking.type=disabled to the five non-thinking Flash
roles so deepseek-v4-flash actually runs in non-thinking mode and
honors the role's temperature/top_p settings:
- simple, simple_json, adviser, searcher, enricher
Pro roles (primary_agent, assistant, generator, refiner, reflector,
coder, installer, pentester) intentionally keep thinking enabled (the
V4 default) for reasoning, tool-use, and security analysis.
PentAGI provider config already supports extra_body as a first-class
yaml field on AgentConfig and forwards it through openai.WithExtraBody,
which the vxcontrol langchaingo fork serializes at the top level of the
Chat Completions request - the same pattern Kimi uses for tool_choice.
No code, schema, or LiteLLM prefix changes required.
Touches:
- backend/pkg/providers/deepseek/config.yml (embedded production config)
- examples/configs/deepseek.provider.yml (user-facing example)
No change to role-to-model mapping, model metadata, pricing, README
wording, LiteLLM prefix, unrelated providers, lifecycle, queues, or
installer flow.
- Update model descriptions to reflect V4 1M context window (up to 384K output)
instead of legacy 128K wording in models.yml and README.
- Split Flash and Pro pricing per official DeepSeek API docs:
- deepseek-v4-flash: input 0.14 / output 0.28 / cache_hit 0.0028 per 1M tokens
- deepseek-v4-pro: input 0.435 / output 0.87 / cache_hit 0.003625 per 1M tokens
- Apply per-role price split across all 13 role configs in both the embedded
config.yml and the user-facing examples/configs/deepseek.provider.yml.
- Replace stale "cache pricing is 10% of input cost" claim in the README,
which no longer holds for either V4 model.
- No change to LiteLLM prefix behavior, role-to-model mapping, lifecycle,
queues, GraphQL schema, migrations, frontend, or installer flow.
The DeepSeek provider config still defaulted to the legacy
`deepseek-chat` and `deepseek-reasoner` model names, which the
upstream DeepSeek API has announced for deprecation on 2026-07-24.
A first-run install therefore breaks once the legacy names are
removed.
Swap the defaults to the current DeepSeek V4 family:
- non-thinking roles use `deepseek-v4-flash`
- reasoning-heavy roles use `deepseek-v4-pro`
The change is limited to the embedded `config.yml` / `models.yml`
inside `backend/pkg/providers/deepseek`, the matching example at
`examples/configs/deepseek.provider.yml`, the `DeepSeekAgentModel`
fallback constant in `deepseek.go`, and three doc references
(README.md, backend/docs/config.md, backend/docs/llms_how_to.md)
plus one installer help string in
`backend/cmd/installer/wizard/locale/locale.go`. LiteLLM prefix
behavior is untouched.
- Included two new provider YAML files for Qwen 3.6 35B models: `vllm-qwen3.6-35b-a3b-fp8-no-think.provider.yml` and `vllm-qwen3.6-35b-a3b-fp8.provider.yml`.
- Updated Dockerfile to copy the new configuration files into the appropriate directory.
- Reword PR #268 references to talk about review feedback rather
than the PR being rejected.
- Clarify host.docker.internal availability: not universally provided
by the core compose stack; Docker Desktop typically resolves it,
while Linux/operator-managed compose stacks may need an explicit
extra_hosts: host.docker.internal:host-gateway entry or another
controlled endpoint.
- Make the safe initial Burp example unambiguously read-only: drop
start_active_scan from the illustrative allowlist and call out
that active capabilities belong to a later, explicitly gated
milestone with scope and approval controls.
- Rephrase the awkward 'PentAGI must not be inferred...' sentence
for clarity.
Signed-off-by: mason5052 <ehehwnwjs5052@gmail.com>
Address Copilot review feedback on PR #306:
- Lifecycle diagram: show running <-> waiting (waiting is a paused state that resumes to running when user input arrives) and document that both running and waiting can reach the terminal statuses finished or failed.
- Replace overloaded 'finished' wording with explicit 'terminal' semantics throughout. finished and failed remain distinct terminal statuses; the queue and webhook layers treat both as terminal.
- Align webhook event names with status terminology: flow.finished for success, flow.failed for failure. Update payload example accordingly and note the failed-flow shape.
Signed-off-by: mason5052 <ehehwnwjs5052@gmail.com>
Proposes a design direction for native flow concurrency control and
completion notifications. The RFC follows the maintainer's relocated
proposal pattern at examples/proposals/<topic>.md and explicitly
builds on the lessons from PR #268 (rejected because the in-memory
queue was hidden lifecycle state).
The RFC covers:
- Goals limited to capping concurrent flows, persisting queued flows
as first-class lifecycle, replacing external polling with at-least-
once webhooks, and preserving the existing createFlow contract.
- Non-Goals that explicitly forbid hidden in-memory queues, multi-
tenant scheduling, generic event bus features, and changing the
meaning of 'finished' for tasks/subtasks/toolcalls.
- Design Principles for persistence, visibility, manageability,
explicit promotion, clear finished semantics, and at-least-once
delivery.
- A proposed concurrency model with a new persisted 'queued' status,
a single MAX_CONCURRENT_FLOWS knob, an explicit promoter, and
full UI/API visibility plus user cancellation.
- A proposed completion webhook model with per-flow and global URLs,
HMAC-SHA256 signatures, persisted deliveries, bounded retries,
and SSRF mitigations.
- Storage and API surface sketches that do not commit to a final
schema.
- Open Questions covering per-user limits, blocking semantics on
createFlow, signature alignment with the issue #235 receipt
direction, and behavior of resources/uploads against queued flows.
- A Suggested First Milestone that lands the queue end-to-end before
webhooks, to keep PR sizes reviewable.
This is documentation only. No runtime code, schema, GraphQL, REST,
or UI behavior changes here.
Refs #298
Signed-off-by: mason5052 <ehehwnwjs5052@gmail.com>
- Added new provider configurations for vLLM Qwen 3.6 in both thinking and non-thinking modes.
- Updated the Dockerfile to include the new configuration files for vLLM Qwen 3.6 and ensure proper setup for deployment.
Dmitry Ng