Commit graph

23 commits

Author SHA1 Message Date
d5b000e18e feat: report what a call cost in tokens
Three dialects spell the same two numbers differently: Ollama counts
in `prompt_eval_count` and `eval_count`, the OpenAI shape in
`usage.prompt_tokens` and `usage.completion_tokens`, Anthropic in
`usage.input_tokens` and `usage.output_tokens`. All three now reach
the caller as `tokens_prompt` and `tokens_completion`, which the hub
lifts into the audit log by itself.

An answer that carries no count leaves both fields empty rather than
reporting a zero: a zero is a measurement, and this is the absence of
one. Half a count is reported as far as it goes.

The cost of a model call was the one part of the audit question
"which model produced this, and what did it cost" that nothing
answered (model pool M3).

Signed-off-by: flemming-it <sf@flemming.it>
2026-10-26 03:54:39 +01:00
Stefan Flemming
a6fe962916 feat: sampling options (temperature, top_p, top_k, seed)
Some checks failed
CI / Linux x86_64 (Forgejo) (push) Failing after 2s
aila asked for these twice (07.08. and 08.08.) and worked around it
by baking the values into a derived Ollama model, which is the only
way they could get a reproducible pipeline.

Verified against a live Ollama: two runs with `temperature: "0"` and
`seed: "42"` return the same answer.

Each wire format gets what it actually supports, and an option the
wire does not have is REFUSED rather than dropped: `top_k` on the
OpenAI wire, `seed` on Anthropic. A silently ignored setting
produces a run that looks configured and is not, which is worse than
an error, especially for `seed`, where the whole point is being able
to repeat the run. A non-numeric value is an error too, so a typo
cannot pass as "server default".

Unset leaves the request byte-identical to 0.2.1, so no existing
caller sees a change.

Eight guards covering all three wires, both refusals, the
unset case, and that `seed` stays a whole number.

Capability 0.2.1 to 0.2.2: four optional inputs, nothing removed.

Signed-off-by: Stefan Flemming <stefan.a.flemming@googlemail.com>
2026-10-16 20:47:40 +02:00
9483f196a1 feat: keep_alive input so a local model stays resident
Some checks failed
CI / Linux x86_64 (Forgejo) (push) Failing after 2s
Ollama unloads a model five minutes after its last request, and the
next request then pays a full reload: 30 to 40 seconds for a large
local model, long enough that an operator assumes something is broken.
The stromnetz demonstrator runs on a notebook from around 12.10. and
hits exactly that after any idle period; it works around it today by
setting OLLAMA_KEEP_ALIVE on the server, which is not expressible in a
flow.

Measured against a live Ollama: without the input the model expired
5:02 after the request, with `keep_alive: "90s"` it expired after 92
seconds. Worth knowing for callers: Ollama keeps the value for the
loaded instance, so a later request that omits keep_alive does not
fall back to five minutes until the model has actually been unloaded.

`-1` is sent as a number rather than a string, because Ollama reads
only the numeric form as "never unload"; quoted it is parsed as a
duration and the setting silently does nothing.

With any other wire format the input is REFUSED, not ignored. It
expresses an expectation about the server, and dropping it quietly
would leave the caller paying the reload while believing it was
handled. The message says what to do instead.

Capability goes 0.2.0 to 0.2.1: the addition is optional and additive,
so `@^0.2` and `@^0` both keep resolving, and an unset value leaves
the request byte-identical to the previous version.

Five guards, including that unset omits the field entirely and that
the other wire formats never carry it. Mutation-checked three ways.

Signed-off-by: flemming-it <sf@flemming.it>
2026-10-07 05:49:16 +02:00
dd7b419d37 chore: cargo fmt
Some checks failed
CI / Linux x86_64 (Forgejo) (push) Failing after 2s
Signed-off-by: flemming-it <sf@flemming.it>
2026-08-20 22:30:33 +02:00
0ec3de9003 fix(ci): embed official.pub — round-trip verify fetched it from a private repo
Some checks failed
CI / Linux x86_64 (Forgejo) (push) Failing after 2s
sign-bundle / sign (push) Successful in 3m20s
The verify step curl-fetched the public key anonymously from the
dev repo's raw URL; since that repo went private the fetch 404s
and every signed release died at verification. Embed the PUBLIC
key verbatim (same pattern as the platform's sync-mirror.sh) so
the workflow is self-contained. Keep in lockstep with
infra/cosign/official.pub on key rotation.

Signed-off-by: flemming-it <sf@flemming.it>
2026-08-20 22:07:15 +02:00
b5ceaa3dfe feat: api accepts 'vllm' as a first-class value
Some checks failed
CI / Linux x86_64 (Forgejo) (push) Failing after 2s
sign-bundle / sign (push) Failing after 1m53s
'openai' reads as the cloud company — an operator running
self-hosted vLLM should be able to write what they mean. The new
value speaks the identical OpenAI wire (a guard test asserts the
emitted request stays byte-identical to api: openai, so the alias
can never drift into a dialect) but yields vLLM-specific error
hints (vllm serve, port 8000) instead of generic OpenAI prose.
Manifest + docs name the value in DE and EN.

Signed-off-by: flemming-it <sf@flemming.it>
2026-08-20 17:13:44 +02:00
065aed4dff fix(ci): finish the fai->chain rename in the sign workflow
Some checks failed
CI / Linux x86_64 (Forgejo) (push) Failing after 2s
sign-bundle / sign (push) Failing after 4m52s
The vendored sign.yml predated the product rename: it installed
and invoked the old CLI name, packed a .fai bundle, and verified
against the renamed fai/platform repo URL — every tag build died
on the missing binary. Align with the platform's current
sign-bundle template: chain CLI, .chain bundle extension,
fai/chain-private raw URL for official.pub.

Signed-off-by: flemming-it <sf@flemming.it>
2026-08-20 16:35:49 +02:00
305ab1cecd feat: OpenAI and Anthropic wire-format adapters via new api input
Some checks failed
CI / Linux x86_64 (Forgejo) (push) Failing after 2s
sign-bundle / sign (push) Has been cancelled
New optional input api: ollama|openai|anthropic (default ollama —
existing flows unchanged).

- openai: OpenAI Chat Completions format (/v1/chat/completions),
  Bearer auth — targets vLLM and compatible self-hosted servers.
- anthropic: Messages API (/v1/messages), x-api-key +
  anthropic-version headers, top-level system field, mandatory
  max_tokens (fixed 4096).
- Audit outputs unchanged: model_endpoint/model_name always set;
  model_digest stays Ollama-only (no digest API on openai/anthropic,
  field is left empty rather than fabricated).
- api_key is only ever placed in auth headers; never in outputs,
  errors, or audit events (verified against the event log).
- No streaming, no tool calls.

Bump module + capability version to 0.2.0. Wire-format unit tests
for request serialization and response parsing against fixed JSON
fixtures; live smoke green on the ollama path and on the openai
path against an OpenAI-compatible local endpoint.

Signed-off-by: flemming-it <sf@flemming.it>
2026-07-11 23:19:32 +02:00
b5389fce45 chore(build): pin chain-module-sdk v0.3.0 and adopt #[chain_module]
SDK 0.3.0 drops the fai-legacy #[fai_module] alias; pin the tag and
flip the attribute, rebuilt module.wasm accordingly.

Signed-off-by: flemming-it <sf@flemming.it>
2026-07-11 23:12:39 +02:00
18a84ef4cb fix: clear Ollama-unreachable error naming the endpoint + fix hint
Some checks failed
CI / Linux x86_64 (Forgejo) (push) Failing after 2s
Signed-off-by: flemming-it <sf@flemming.it>
2026-07-07 19:18:24 +02:00
88c5a0c158 chore(build): rebuild module.wasm against current module ABI
Prebuilt wasm exported a stale ABI (no exported instance
chain:platform/invoke@1.0.0); rebuilt from source so llm.chat loads in
the current hub.

Signed-off-by: flemming-it <sf@flemming.it>
2026-07-06 01:22:12 +02:00
38f9eeac0f chore: org rename fai-modules/fai-plugins -> chain-modules/chain-plugins + LICENSE holder
Some checks failed
CI / Linux x86_64 (Forgejo) (push) Failing after 2s
Signed-off-by: flemming-it <sf@flemming.it>
2026-06-16 18:25:36 +02:00
f7ec4e59bf docs/ci: chain-*-sdk references (renamed SDK)
Some checks failed
CI / Linux x86_64 (Forgejo) (push) Failing after 1s
Signed-off-by: flemming-it <sf@flemming.it>
2026-06-16 11:32:22 +02:00
d5894c7bd4 refactor: depend on renamed chain-module-sdk
Some checks failed
CI / Linux x86_64 (Forgejo) (push) Failing after 1s
Signed-off-by: flemming-it <sf@flemming.it>
2026-06-16 11:28:12 +02:00
d5b4d339ce chore: contact email platform@ -> chain@flemming.ai
Some checks failed
CI / Linux x86_64 (Forgejo) (push) Failing after 2s
Signed-off-by: flemming-it <sf@flemming.it>
2026-06-16 10:16:30 +02:00
782a48cee2 docs: old product name F∆I Platform -> Ch∆In
Some checks are pending
CI / Linux x86_64 (Forgejo) (push) Waiting to run
The product is Ch∆In; the old 'F∆I Platform' name survived in module
docs/manifests. Generic 'platform' terms left untouched.

Signed-off-by: flemming-it <sf@flemming.it>
2026-06-16 10:15:26 +02:00
dc8cab7813 refactor: rename to chain:platform WIT + provider chain
Some checks failed
CI / Linux x86_64 (Forgejo) (push) Failing after 2s
Tracks the platform's fai->chain rename: capability provider chain,
WIT ABI chain:platform (binding paths/accessors), SDK pin bumped.

Signed-off-by: flemming-it <sf@flemming.it>
2026-06-15 22:54:53 +02:00
e595d8e328 feat: schema_version 3 with DE/EN field descriptions
Some checks failed
CI / Linux x86_64 (Forgejo) (push) Failing after 1s
Editor renders the description in the active locale as a
tooltip on each port. EN remains the default; DE peer
provided for every field.

Signed-off-by: flemming-it <sf@flemming.it>
2026-06-01 22:13:17 +02:00
e0acb69524 ci: sign release bundles with FAI_SIGNING_KEY
Some checks failed
CI / Linux x86_64 (Forgejo) (push) Failing after 2s
Drop-in from fai/platform .forgejo/workflow-templates/
sign-bundle-keypair.yml. Triggers on v*.*.* tag push: builds
wasm32-wasip2, fetches the fai CLI from get.fai.flemming.ai,
packs the bundle, ECDSA P-256 signs it against the org-level
Forgejo secret FAI_SIGNING_KEY, round-trip-verifies against
infra/cosign/official.pub on fai/platform main, and attaches
bundle + .sig to the Forgejo Release via the API.

Signed-off-by: flemming-it <sf@flemming.it>
2026-05-28 23:14:20 +02:00
05660a4e01 chore(module): schema_version 2 + explicit provider
Some checks failed
CI / Linux x86_64 (Forgejo) (push) Failing after 2s
Brings module.yaml in line with F∆I 0.12.0's capability naming
convention. The legacy schema_version 1 (with implicit provider)
still loads in 0.12.0 via the grace-period default, but this
bump makes the publisher identity explicit and unblocks
storage-layout migration via 'fai migrate'.

See fai/platform docs/reference/capability-namespaces.yaml +
docs/operations/migration-to-0.12.md.

Signed-off-by: flemming-it <sf@flemming.it>
2026-05-28 11:48:25 +02:00
481c0c209f chore: SDK URL → git.flemming.ai/fai/module-sdk-rust
Some checks failed
CI / Linux x86_64 (Forgejo) (push) Failing after 2s
The fai-module-sdk repo was renamed from `fai/module-sdk` to
`fai/module-sdk-rust` to align with the three SDK families
convention (base + language suffix). Pin updated accordingly.
The old git.flemming.ws URL is also retired in this pass.

Signed-off-by: flemming-it <sf@flemming.it>
2026-05-26 13:16:53 +02:00
84a23e6b14 docs: add MODULE.md + MODULE.de.md inline docs
Some checks failed
CI / Linux x86_64 (Forgejo) (push) Failing after 1s
Explains the 'why this module instead of inline HTTP' answer
operators ask when they see llm.chat for the first time:
audit, permission posture, endpoint portability.

Signed-off-by: flemming-it <sf@flemming.it>
2026-05-25 14:06:04 +02:00
4c10a5ff3d feat: initial llm-chat v0.1.0 (llm.chat@0.1.0)
All checks were successful
CI / Linux x86_64 (Forgejo) (push) Successful in 1m57s
Generic Ollama-compatible LLM chat adapter. The lower-level
counterpart to orchestrator-llm: orchestrator-llm wraps the LLM
in a planning prompt that emits a structured F∆I Plan; this
module is the plain-prompt adapter that flows compose for
summarisation, translation, free-form Q&A, etc.

Capability surface:

  Inputs:
    prompt        : text
    endpoint      : text   (Ollama /api/chat URL)
    model         : text
    api_key       : text   (optional bearer token)
    system_prompt : text   (optional)

  Outputs:
    response       : text   (assistant reply)
    model_endpoint : text   (audit correlation)
    model_name     : text   (audit correlation)
    model_digest   : text   (Ollama /api/show probe; empty
                              for non-Ollama or transient failures)

  Permissions: net to localhost / 127.0.0.1 / api.openai.com /
               api.anthropic.com.

The audit-field trio (endpoint + name + digest) closes the same
forensic gap that orchestrator-llm v0.3.1 closed for plan
generation: any historical chat invocation can be traced to the
exact model that produced it.

Implementation reuses the same defensive Ollama client pattern
from orchestrator-llm — derive_show_url + extract_show_digest
+ best-effort probe_model_digest. Duplication accepted at the
two-module mark; a shared crate refactor lands once a third
module needs the same plumbing.

12 host-side tests cover prompt building, Ollama-shaped
response parsing, URL transform, digest extraction (top-level
+ nested), end-to-end success, end-to-end probe-failure
swallow, end-to-end skip-for-non-Ollama, and the missing-input
guards.

Wasm artifact: 294 KB. Verified to build with v1.0 fai:platform
imports baked in.

Bootstrapped via 'fai new module llm.chat' (workspace v0.10.13)
which now produces an SDK-based template directly.

Signed-off-by: flemming-it <sf@flemming.it>
2026-05-03 23:17:36 +02:00