im UdZddlZddlZddlZddlmZddlmZmZm Z m Z ddl m Z ddl Z ddlZddlmZejeZehdZeeed<ejd ejZd ed efd Ziaeeeeeffed <daeed<dZiZ eeeeeeefffed<iZ!eeefed<dZ"gdZ#e#dZ$dZ%iddddddddddddddddd dd!d"d#d$d%d$d&dd'd(d)d$d*d+d,d-id.d-d/d0d1d2d3d$d4d0d5dd6d7d8d0d9d:d;d<d=d-d>d?d@d2dAd?dBd?dCd-dDd0idEd0dFd0dGd7dHd7dId7dJd0dKd0dLdMdNd7dOd7dPd:dQd-dRddSd-dTd-dUd<Z&dVZ'dWZ(dXZ)dYZ*dZed efd[Z+dZed e,fd\Z-dZed e,fd]Z.id^d_d`d_dadbdcdddedfdgdhdidfdjdkdld9dmdndodndpdqdrdsdtd*dudvdwd3dxdydydzd{d|d}d}d~dZ/eeefed<dZed e efdZ0dZed e,fdZ1dZed e,fdZ2dZed e efdZ3defdZ4ddede5de5d e e5fdZ6deeefde7edfd e e5fdZ8deeefd e e5fdZ9deeefd e e5fdZ:deeefd eeeffdZ;deeeeeffdedeeefd dfdZd efdZ?d eee5ffdZ@d edZede5d dfdZAd edZed e e5fdZBde5d e e5fdZCded e e5fdZDded e e5fdZEdeded e,fdZFd edZed e e5fdZGd edZed e e5fdZHd ed efdZId edZeded e e5fdZJd ed e e5fdZK dd edZedede5dzded e5f dZLded e5fdZMde eeefd e5fdZNdddde eeefdede e eeefd e5fdZOdS)zModel metadata, context lengths, and token estimation utilities. Pure utility functions with no AIAgent dependency. Used by ContextCompressor and run_agent.py for pre-flight context checks. N)Path)AnyDictListOptional)urlparse)OPENROUTER_MODELS_URL>:x-aix.aiz-aiz.aikimi-cnarcee-ai deep-seek ai-gateway minimax-cn copilot-acp moonshot-cn qwen-portal xiaomi-mimo openai-codex opencode-zen github-models google-geminigithub-copilotgoogle-ai-studiogoglmxaizaizengrokkilokimimimonousqwenarceelocalzhipualiyunclaudecustomgeminigithubgoogleollamavercelxiaomialibabaarceeaicopilotminimaxdeepseekkilocodemoonshotopencode anthropic dashscope openrouter qwen-oauth kimi-coding opencode-go ollama-cloudkimi-coding-cn_PROVIDER_PREFIXESzE^(\d+\.?\d*b|latest|stable|q\d|fp?\d|instruct|chat|coder|vision|text)modelreturnc,d|vs|dr|S|dd\}}|}|tvr0t |r|S|S|S)uaStrip a recognised provider prefix from a model string. ``"local:my-model"`` → ``"my-model"`` ``"qwen3.5:27b"`` → ``"qwen3.5:27b"`` (unchanged — not a provider prefix) ``"qwen:0.5b"`` → ``"qwen:0.5b"`` (unchanged — Ollama model:tag) ``"deepseek:latest"``→ ``"deepseek:latest"``(unchanged — Ollama model:tag) :http) startswithsplitstriplowerrD_OLLAMA_TAG_PATTERNmatch)rEprefixsuffix prefix_lowers >''))L)))  $ $V\\^^ 4 4 L L_model_metadata_cache_model_metadata_cache_timei_endpoint_model_metadata_cache#_endpoint_model_metadata_cache_timei,)i}i>i@r\zclaude-opus-4-7i@Bzclaude-opus-4.7zclaude-opus-4-6zclaude-sonnet-4-6zclaude-opus-4.6zclaude-sonnet-4.6r,i@ z gpt-5.4-nanoiz gpt-5.4-minizgpt-5.4izgpt-5.3-codex-sparkr[z gpt-5.1-chatzgpt-5zgpt-4.1izgpt-4r.iz gemma-4-31biz gemma-4-26bzgemma-3igemmai r8llamazqwen3-coder-plusz qwen3-coderir'r7i rizgrok-code-fastz grok-4-1-fastiz grok-2-visionz grok-4-fastz grok-4.20zgrok-4zgrok-3zgrok-2r"r$trinityelephantzQwen/Qwen3.5-397B-A17BzQwen/Qwen3.5-35B-A3Bzdeepseek-ai/DeepSeek-V3.2izmoonshotai/Kimi-K2.5zmoonshotai/Kimi-K2-ThinkingzMiniMaxAI/MiniMax-M2.5zXiaomiMiMo/MiMo-V2-Flashz mimo-v2-proz mimo-v2-omniz mimo-v2-flashz zai-org/GLM-5) context_lengthcontext_windowmax_context_lengthmax_position_embeddings max_model_lenmax_input_tokensmax_sequence_length max_seq_len n_ctx_trainn_ctx)max_completion_tokensmax_output_tokens max_tokens) localhostz 127.0.0.1z::1z0.0.0.0)z.docker.internalz.containers.internalz.lima.internalbase_urlcT|pddS)N/)rMrstripros rT_normalize_base_urlrus& N ! ! # # * *3 / //rVcHdt|vS)N openrouter.ai)rurNrts rT_is_openrouter_base_urlrxs" 1(;;AACC CCrVc^t|}t|ot| SN)ruboolrx)ro normalizeds rT_is_custom_endpointr}s.$X..J    G$;J$G$G GGrVzapi.openai.comopenaiz chatgpt.comapi.anthropic.comr<zapi.z.air zapi.moonshot.air@zapi.moonshot.cnrCz api.kimi.comz api.arcee.air(z api.minimaxzdashscope.aliyuncs.comr4zdashscope-intl.aliyuncs.comzportal.qwen.air?rwr>z!generativelanguage.googleapis.comzinference-api.nousresearch.comr&zapi.deepseek.comzapi.githubcopilot.comr6 fireworksrArr3rB)zmodels.github.aizapi.fireworks.aiz opencode.aizapi.x.aizapi.xiaomimimo.comzxiaomimimo.comz ollama.com_URL_TO_PROVIDERct|}|sdStd|vr|nd|}|jp|j}t D] \}}||vr|cSdS)aInfer the models.dev provider name from a base URL. This allows context length resolution via models.dev for custom endpoints like DashScope (Alibaba), Z.AI, Kimi, etc. without requiring the user to explicitly set the provider name in config. N://zhttps://)rurnetlocrNpathritems)ror|parsedhosturl_partproviders rT_infer_provider_from_urlrs%X..J t EZ$7$7jj=T =T=T U UF =   7FK$5$5$7$7D.4466( t  OOO  4rVc$t|duSrz)rrts rT_is_known_provider_base_urlr s #H - -T 99rVc t|}|sdSd|vr|nd|} t|}|jpd n#t$rYdSwxYw tvrdSt fdt DrdSddl} | }|j p |j p|j S#t$rYnwxYw d }t|d krj t|dt|d }}|d krdS|d krd|cxkrdkrnndS|dkr|dkrdSn#t$rYnwxYwdS)zOReturn True if base_url points to a local machine (localhost / RFC-1918 / WSL).Frzhttp://rqTc3BK|]}|VdSrz)endswith).0rRrs rT z$is_local_endpoint..s/ I IV4== I I I I I IrVrN.rJ )rurhostname Exception _LOCAL_HOSTSany_CONTAINER_LOCAL_SUFFIXES ipaddress ip_address is_private is_loopback is_link_local ValueErrorrLlenint) ror|urlrraddrpartsfirstsecondrs @rTis_local_endpointrs$X..J u++**1G:1G1GC#$" uu |t I I I I/H I I IIIt ##D))H$"2Hd6HH       JJsOOE 5zzQ aMM3uQx==6E{{t||f 2 2 2 2 2 2 2 2 2t||# t    D  5s?: AA=)B'' B43B4 0D9D9* D99 EEcddl}t|}|}|dr |dd} |d5} ||d}|jdkr dddd Sn#t $rYnwxYw ||d }|jdkr8 |}d |vr dddd Sn#t $rYnwxYwn#t $rYnwxYw ||d }|jdkr||d}|jdkrd|jvr ddddSn#t $rYnwxYw ||d}|jdkr&|}d|vr ddddSn#t $rYnwxYwdddn #1swxYwYn#t $rYnwxYwdS)zDetect which local server is running at base_url by probing known endpoints. Returns one of: "ollama", "lm-studio", "vllm", "llamacpp", or None. rN/v1g@timeout/api/v1/models lm-studioz /api/tagsmodelsr1 /v1/props/propsdefault_generation_settingsllamacppz/versionversionvllm) httpxrurClientget status_coderjsontext)rorr| server_urlclientrdatas rTdetect_local_server_typer6sQ LLL$X..JJ5!!%_ * \\#\ & &' & JJ*<<<===C''& ' ' ' ' ' ' ' ' (     JJ*77788=C'' vvxx#t++#+#' ' ' ' ' ' ' '  ,$     JJ*77788=C'' j#8#8#899A=C'',IQV,S,S%9' ' ' ' ' ' ' ' :     JJ*66677=C''6688D D((%K' ' ' ' ' ' ' ' L    M' ' ' ' ' ' ' ' ' ' ' ' ' ' ' P       4sG G#B2 G?G B  G B  G#C-5C GC- C)&C-(C))C-,G- C:7G9C::G>AE GG E)&G(E))G-;F7) G6G7 GGGG GGGGG G,+G,valuec#Kt|tr2|V|D]}t|Ed{VdSt|tr|D]}t|Ed{VdSdSrz) isinstancedictvalues_iter_nested_dictslist)rnesteditems rTrrqs%0 llnn 2 2F)&11 1 1 1 1 1 1 1 1 2 2 E4 0 0 0D)$// / / / / / / / /00 0 0rV逖minimummaximumc$ t|trdSt|tr(|dd}t |}n#t tf$rYdSwxYw||cxkr|krnn|SdS)N,rq)rr{strrMreplacer TypeErrorr)rrrresults rT_coerce_reasonable_intr{s eT " " 4 eS ! ! 3KKMM))#r22EU z "tt&####G##### 4sA&A A&&A;:A;payloadkeys.cd|D}t|D]W}|D]@\}}t||vr)t |}||ccSAXdS)Nc6h|]}|S)rN)rkeys rT z%_extract_first_int..s * * *cciikk * * *rV)rrrrNr)rrkeysetmappingrrcoerceds rT_extract_first_intrs * *T * * *F%g..!--//  JC3xx~~v--,U33G"#   4rVc,t|tSrz)r_CONTEXT_LENGTH_KEYSrs rT_extract_context_lengthr g'; < <z$_extract_pricing..s.PPP*#uc#hhnn&&PPPrVc3NK|]}tfd|DV dS)c3 K|]}|vV dSrzr)raliasr|s rTrz-_extract_pricing...s(@@uu *@@@@@@rVN)r)raliasesr|s rTrz#_extract_pricing..s@ccW3@@@@@@@@@ccccccrV)Nrq)rrrr)r alias_maprpricingtargetrrr|s @rT_extract_pricingr sR`.fk I&g..  PP PPP ccccPYP`P`PbPbccccc  "$(00  OFG   J&&:e+2)F)F)J)JKbdh)i)i &(33 99Y33 E uh 6 6 6 "2B77I .sVJq$4G4Gj!!Z/rVrqrrrrj model_aliasz1Failed to fetch model metadata from %s/models: %s)rurxrYrrZr_ENDPOINT_MODEL_CACHE_TTLrrsappendrrrrrrrr rrrokrrr)ror%rr|cached cached_at candidates alternater) last_error candidaterr!rr rEr rrarkr  is_llamacppbase props_respprops gen_settingsrjr.excs rTfetch_endpoint_model_metadatar>s/%X..J 0<< /33J??7;;JJJ  49;;#:>W"W"WMJ5!!'ssO**3// & %Yj00)$$$8?G 3' 3 344RG&*J33 s##i/1 |C"EEEH  % % ' ' 'mmooG/1E VR00 ; ;!%.. 99T??)/681L1L(M!8!?!?!-.Bikk / ;LLL   JJJJJJ b H*V`aaa13":.6:ikk' 3 Is>>D"L.!CK54L.5 L?L.L)L.. M8MMc(ddlm}|dz S)z8Return path to the persistent context length cache file.rget_hermes_homezcontext_length_cache.yaml)hermes_constantsrAr@s rT_get_context_cache_pathrC:s(000000 ?  : ::rVcZt}|siS t|5}tj|pi}dddn #1swxYwY|diS#t $r'}td|icYd}~Sd}~wwxYw)z:Load the model+provider -> context_length cache from disk.Ncontext_lengthsz'Failed to load context length cache: %s) rCexistsopenyaml safe_loadrrrr)rfrr#s rT_load_context_cacherK@s " $ $D ;;==  $ZZ +1>!$$*D + + + + + + + + + + + + + + +xx)2...  >BBB s@A9A A9AA9A A99 B*B%B*%B*lengthc|d|}t}|||krdS|||<t} |jddt |d5}t jd|i|ddddn #1swxYwYt d ||d dS#t$r&}t d |Yd}~dSd}~wwxYw) zPersist a discovered context length for a model+provider combo. Cache key is ``model@base_url`` so the same model name served from different providers can have different limits. @NT)parentsexist_okwrEF)default_flow_stylez%Cached context length %s -> %s tokensrz'Failed to save context length cache: %s) rKrrCparentmkdirrGrHdumprinforr)rErorLrr rrJr#s rTsave_context_lengthrWNsi   X  C  ! !E yy~~E#J " $ $DC $666 $__ O I(%0! N N N N O O O O O O O O O O O O O O O ;SV--PPPPP CCC >BBBBBBBBBCs<,C1B CBCB!C C2 C--C2cV|d|}t}||S)zBLook up a previously discovered context length for model+provider.rN)rKr)rErorr s rTget_cached_context_lengthrYcs0   X  C  ! !E 99S>>rVcurrent_lengthc.tD] }||kr|cS dS)z@Return the next lower probe tier, or None if already at minimum.N)CONTEXT_PROBE_TIERS)rZtiers rTget_next_probe_tierr^js/# . KKK ! 4rV error_msgc|}gd}|D]O}tj||}|r6t|d}d|cxkrdkrnK|cSPdS)a?Try to extract the actual context limit from an API error message. Many providers include the limit in their error text, e.g.: - "maximum context length is 32768 tokens" - "context_length_exceeded: 131072" - "Maximum context size 32768 exceeded" - "model's max context length is 65536" )zY(?:max(?:imum)?|limit)\s*(?:context\s*)?(?:length|size|window)?\s*(?:is|of|:)?\s*(\d{4,})z:context\s*(?:length|size|window)\s*(?:is|of|:)?\s*(\d{4,})z)(\d{4,})\s*(?:token)?\s*(?:context|limit)z">\s*(\d{4,})\s*(?:max|limit|token)z(\d{4,})\s*(?:max(?:imum)?)\brJrrNrNresearchrgroup)r_ error_lowerpatternspatternrPlimits rTparse_context_limit_from_errorrirs//##KH ';//   A''Eu**** ***** 4rVc|}d|vod|vpd|v}|sdSgd}|D]E}tj||}|r,t|d}|dkr|cSFdS)u#Detect an "output cap too large" error and return how many output tokens are available. Background — two distinct context errors exist: 1. "Prompt too long" — the INPUT itself exceeds the context window. Fix: compress history and/or halve context_length. 2. "max_tokens too large" — input is fine, but input + requested_output > window. Fix: reduce max_tokens (the output cap) for this call. Do NOT touch context_length — the window hasn't shrunk. Anthropic's API returns errors like: "max_tokens: 32768 > context_window: 200000 - input_tokens: 190000 = available_tokens: 10000" Returns the number of output tokens that would fit (e.g. 10000 above), or None if the error does not look like a max_tokens-too-large error. rmavailable_tokenszavailable tokensN)zavailable_tokens[:\s]+(\d+)zavailable\s+tokens[:\s]+(\d+)z =\s*(\d+)\s*$rJra)r_reis_output_cap_errorrfrgrPtokenss rT(parse_available_output_tokens_from_errorrns //##K  # U ; . S2D 2S tH  ';//  Q((F{{ 4rV candidate_id lookup_modelcb||krdSd|vr"|ddd|krdSdS)aReturn True if *candidate_id* (from server) matches *lookup_model* (configured). Supports two forms: - Exact match: "nvidia-nemotron-super-49b-v1" == "nvidia-nemotron-super-49b-v1" - Slug match: "nvidia/nvidia-nemotron-super-49b-v1" matches "nvidia-nemotron-super-49b-v1" (the part after the last "/" equals lookup_model) This covers LM Studio's native API which stores models as "publisher/slug" while users typically configure only the slug after the "local:" prefix. TrrrJF)rsplit)rorps rT_model_id_matchesrssH|##t l|223::1=MMt 5rVcddl}t|}|d}|dr |dd} t |}n#t $rYdSwxYw|dkrdS |d5}||d d |i }|jd kr ddddS| }| d d} d| vr| dD]s} d| vrm| } t| dkr4 t| dccdddS#t$rYowxYwt| di} | D]B\} }d| vr9t#|tt$frt|ccdddSC dddn #1swxYwYn#t $rYnwxYwdS)ayQuery an Ollama server for the model's context length. Returns the model's maximum context from GGUF metadata via ``/api/show``, or the explicit ``num_ctx`` from the Modelfile if set. Returns None if the server is unreachable or not Ollama. This is the value that should be passed as ``num_ctx`` in Ollama chat requests to override the default 2048. rNrrrrr1@r /api/showrrr parametersrqnum_ctx  model_infora)rrUrsrrrrpostrrrrLrMrrrrrfloat)rErorrr server_typerresprparamslinerr}rrs rTquery_ollama_num_ctxrsLLL'..J%%J5!!%_ .x88 ttht \\#\ & & &&;;*777vz>R;SSD3&& & & & & & & & &99;;DXXlB//FF"""LL..%%D D(( $ 2 2 4 4u::??%'*59~~ 5 5 & & & & & & & &$.%%% $%,33J(..00 & & U#s**z%#u/N/N*u::%%- & & & & & & & &( &) & & & & & & & & & & & & & & &.      4s A A'&A'3G< )G02 G<?BG0E%G0 G<% E2/G01E22A#G0 G<"G0$ G<0G44G<7G48G<< H H cR ddl}t|}|d}|dr |dd} t |}n#t $rd}YnwxYw |d5}|dkrL||d d |i }|jd kr$| }| d d}d|vr| dD]s} d| vrm| } t| dkr4 t| dccdddS#t$rYowxYwt| di} | D]B\} } d| vr9t#| tt$frt| ccdddSC|dkr| |d}|jd kr]| }| dgD]1}t'| dd|s$t'| dd|r| dgD]j}| di}| d}|r;t#|tt$frt|cccdddSk| dp| d}|r9t#|tt$frt|ccdddS3| |d|}|jd kr| }| dp)| dp| d}|r7t#|tt$frt|cdddS| |d }|jd kr| }| d!g}|D]}t'| dd|rz| dp)| dp| d}|r9t#|tt$frt|ccdddSdddn #1swxYwYn#t $rYnwxYwdS)"z4Query a local server for the model's context length.rNrrrrrurr1rvrrwrrxrqryrzr{r|r}rarrrrrloaded_instancesconfigrcz /v1/models/rermz /v1/modelsr)rrUrsrrrrr~rrrrLrMrrrrrrrs)rErorrrrrrrrrr}rrr,instcfgctx models_lists rT_query_local_context_lengthrsLLL #5 ) )E%%J5!!%_ .x88  H \\#\ & &E ,&h&&{{j#;#;#;65/{RR#s**99;;D"XXlB77F F**$*LL$6$6--D(D00(, (:(:(<(<#&u::??%-/259~~(=(='E ,E ,E ,E ,E ,E ,E ,E ,(,6%-%-%-(,%-"&,!;!;J&0&6&6&8&8.. U+s22z%#u7V7V2#&u::--5E ,E ,E ,E ,E ,E ,E ,E ,Bk))zzZ"?"?"?@@#s**99;;D!XXh33 0 0,QUU5"-=-=uEE 0IZ[\[`[`aegi[j[jlqIrIr 0().@"(E(E44&*hhx&<&<&)gg.>&?&?#&!4:cC<+H+H!4+.s88OOOOYE ,E ,E ,E ,E ,E ,E ,E ,\#$%%(<"="="XGWAXAXC"0z#U|'D'D0'*3xxaE ,E ,E ,E ,E ,E ,E ,E ,f::????@@D3&&yy{{hh//g488N8O8OfSTSXSXYeSfSf,:cC<#@#@,#&s88OOKE ,E ,E ,E ,E ,E ,E ,E ,E ,E ,E ,E ,E ,E ,E ,E ,E ,E ,E ,E ,E ,E ,E ,L       4s A A('A(,TB2T 5E T  T E%"T $E%%A#T  TDT ) T6AT  TB&T  TC"T 1 T>T ? T TTTT T$#T$c.|ddS)zNormalize version separators for matching. Nous uses dashes: claude-opus-4-6, claude-sonnet-4-5 OpenRouter uses dots: claude-opus-4.6, claude-sonnet-4.5 Normalize both to dashes for comparison. r-)r)rEs rT_normalize_model_versionr`s ==c " ""rVcX|r|drdS |d}|dr |dd}|d}|dd}tj||d }|jd krdS|}|d gD]O}|d |kr4|d} t| tr | dkr| cSPn2#t$r%} t d| Yd} ~ nd} ~ wwxYwdS)zQuery Anthropic's /v1/models endpoint for context length. Only works with regular ANTHROPIC_API_KEY (sk-ant-api*). OAuth tokens (sk-ant-oat*) from Claude Code return 401. z sk-ant-oatNrrrrz/v1/models?limit=1000z 2023-06-01)z x-api-keyzanthropic-versionrr(rrrrfrz%Anthropic /v1/models query failed: %s) rKrsrrrrrrrrrr) rEror%r9rr)rrr,rr#s rT_query_anthropic_context_lengthrjsq g((66tAs## ==   9D,,, !-  |C"===  s " "4yy{{&"%%  AuuT{{e##ee.//c3''C!GGJJJ   AAA r s 3 3 Iu##E** 3I#e**,, #e**0E0N0Nyy!12222222  3 4rVconfig_context_lengthrc|t|tr|dkr|St|}|rt||}||St |rBt |s2t ||}||}|sht|dkr/tt| }n&| D]\}} ||vs||vr| }n|r,|d} t| tr| St |set|r+t||} | r| dkrt||| | St d||t$dt$S|dks|rd |vrt'||pd |} | r| S|d ks|r&d |vr" dd lm} | |S#t,$rYnwxYw|}|r|dvr|rt/|}|r|}|dkrt1|} | r| S|rddlm}|||} | r| St7}||vr||ddS|}t;t< ddD] \}}||vr|cS|r:t|r+t||} | r| dkrt||| | St$S)acGet the context length for a model. Resolution order: 0. Explicit config override (model.context_length or custom_providers per-model) 1. Persistent cache (previously discovered via probing) 2. Active endpoint metadata (/models for explicit custom endpoints) 3. Local server query (for local endpoints) 4. Anthropic /v1/models API (API-key users only, not OAuth) 5. OpenRouter live API metadata 6. Nous suffix-match via OpenRouter cache 7. models.dev registry lookup (provider-aware) 8. Thin hardcoded defaults (broad family patterns) 9. Default fallback (128K) Nr)r%rJrauCould not detect context length for model %r at %s — defaulting to %s tokens (probe-down). Set model.context_length in config.yaml to override.rr<rzhttps://api.anthropic.combedrockzbedrock-runtime)get_bedrock_context_length)r>r-r&)lookup_models_dev_contextr[c,t|dS)Nrr)xs rTz*get_model_context_length..&ss1Q4yyrVT)rreverse)rrrUrYr}rr>rrnextiterrrrrrWrrVDEFAULT_FALLBACK_CONTEXTragent.bedrock_adapterr ImportErrorrragent.models_devrr$rNsortedDEFAULT_CONTEXT_LENGTHS)rEror%rrr2endpoint_metadatamatchedrrra local_ctxrreffective_providerinferredrrr default_modelrLs rTget_model_context_lengthrs,(Z8Ms-S-S(XmpqXqXq$$ #5 ) )E*5(;;  M8$$,-H-R-R,9(GTTT#''.. $%%**t$5$<$<$>$>??@@#4"9"9";";JC||se||"'(4  &$[[)9::N.#.. &%%*844 , ** %7xHH %Q'xCCC$$ KK.x$>>>>>''(:EBB  J$%%H ""#3V<<< ++--K!'%%''-@-@$""" v K ' 'MMM (%h///x@@  Q x ; ; ;  $#sG G%$G%rc4|sdSt|dzdzS)aRough token estimate (~4 chars/token) for pre-flight checks. Uses ceiling division so short texts (1-3 chars) never estimate as 0 tokens, which would cause the compressor and pre-flight checks to systematically undercount when many short tool results are present. rrr)rs rTestimate_tokens_roughr6s& q IIMa rVmessagescDtd|D}|dzdzS)z:Rough token estimate for a message list (pre-flight only).c3NK|] }tt|V!dSrzrrrmsgs rTrz1estimate_messages_tokens_rough..Ds.88c#c((mm888888rVrr)sum)r total_charss rTestimate_messages_tokens_roughrBs-88x88888K !O !!rV) system_prompttoolsrrcd}|r|t|z }|r|td|Dz }|r|tt|z }|dzdzS)u<Rough token estimate for a full chat-completions request. Includes the major payload buckets Hermes sends to providers: system prompt, conversation messages, and tool schemas. With 50+ tools enabled, schemas alone can add 20-30K tokens — a significant blind spot when only counting messages. rc3NK|] }tt|V!dSrzrrs rTrz0estimate_request_tokens_rough..Ys.==S3s3xx========rVrr)rrr)rrrrs rTestimate_request_tokens_roughrHsyK*s=))) >s==H======  's3u::& !O !!rV)rr)F)rqF)rqrqNrq)P__doc__rrbrpathlibrtypingrrrr urllib.parserrrHrBr getLogger__name__r frozensetrDr__annotations__compile IGNORECASErOrUrWrXrrrYrZr/r\rMINIMUM_CONTEXT_LENGTHrrrrrrur{rxr}rrrrrrrrtuplerrrr rr$r>rCrKrWrYr^rirnrsrrrrrrrrrrrVrTrsS   ,,,,,,,,,,,,!!!!!! 222222  8 $ $ &/Y000&&IcN*!bjLM ##&46tCc3h/0555$%E%%%GIS$sDcN/B*C%C DIII8:#T#u*%5::: /q1   M w M w MwMMwMM fMFM F!M"w#M$6%M&F'M( V)M*w+M, V-M0 g1M465MM667M8v9M: T;M>?MB VCMHIMJ6KML FMMRvSMV 6WMdfeMfWgMhTiMj7kMlmMn foMp fqMMMr fsMt FuMx FyM|v}M@AMDfEMFFGMH IMJFKML"6MMNfOMPQMR7SMTFUMVVWMXVYMM^ < 0#0#0000DcDdDDDDH#H$HHHH $h$8$$ $ } $ ' $M$G$9$i$"9$l$\$($%f$  !$"Y#$$"# " 1$$$$sCx.8sx}$:#:$::::$$$$$$N8s8x}8888v0c0000  #  S ZbcfZg     S#X eCHo (SV-    =T#s(^= =====DcN=x}====d38nc3h.,d3S#X#67,3,tTWY\T\~,bf,,,,!+!+!+c4S>>Q9R!+!+!+!+LXXX XX #tCH~  XXXXv;;;;; T#s(^    CsCcC3C4CCCC*SCHSM chsm8(( ((((VCst&22s2x}2222j\s\c\hsm\\\\~#C#C####3#PXY\P]> F(, I$I$ I$I$I$: I$  I$  I$I$I$I$X         "T$sCx.-A"c"""",0 """4S>""" Dc3h( ) "  """"""rV